Development and application of control concepts for twin-screw wet granulation in the ConsiGmaTM-25, Part 1: Granule composition

Continuous manufacturing of pharmaceuticals offers several benefits, such as increased production efficiency, enhanced product quality control, and lower environmental footprint. To fully exploit these benefits, standard operation mode (production processes with no or minimal disturbance mitigation measures) should be supported by adopting novel quality-by-control (QbC) methodologies. The paper at hand is the first part of a study focused on developing QbC algorithms for optimizing twin-screw wet granulation in the industrial manufacturing line ConsiGmaTM-25, specifically addressing granule composition. This work relieson previously established process-analytical-technology (PAT) equipment for real-time monitoring of the granule composition, i.e., the active pharmaceutical ingredient (API) and liquid content in wet granules. The developed control platform integrates model-based process control algorithms that aim to keep the API- and liquid content at target values through real-time adjustments of the process parameters. Furthermore, the platform integrates a digital operator assistant, which aims to detect and classify granulation disturbances and provides messages and instructions for the plant operator. The present manuscript systematically outlines all design steps from the development phase in the simulation environment to the final real system application and validation. The control platform's performance is demonstrated through selected test scenarios on the ConsiGmaTM-25 manufacturing line. The obtained results indicate improved disturbance robustness and an increase in intermediate/final product quality (compared to conventional operating modes): The process control algorithms successfully maintained the API- and liquid content at target values despite process disturbances. Furthermore, realistic disturbances (feeder, pump, and material) were accurately detected and classified by the digital assistant algorithm. The information was provided through a user interface, offering real-time support for plant personnel.

Optimal detection and classification of grid connected system using MSVM-FSO technique

This paper, a hybrid method, is proposed for protecting the hybrid photovoltaic (PV) and wind turbine (WT) system. The proposed protecting method is the hybrid wrapper of both the multiple support vector machine (MSVM) and firebug swarm optimization (FSO), commonly named as MSVM-FSO method. The proposed technique is diagnosing the appropriate fault occurring in the hybrid system. The main purpose of the proposed system is to assure the system with lower complexity for the fault diagnosis and detection (FDD) for improving the power quality (PQ) of hybrid method. Here, the MSVM approach is used to detect the fault conditions of grid-tied system. To evaluate the events of voltages, fault and the currents of hybrid systems are analyzed at the feeder of buses. The FSO categorizes the types of fault, which is occurred in grid-connected system. By then, the proposed method's performance is done in the MATLAB software and it is contrasted with different existing methods. From this, the proposed method provides accuracy as 99.7% and efficiency as 98%, which is high compared to existing methods.

Asynchronous adaptive event-triggered fault detection for delayed Markov jump neural networks: A delay-variation-dependent approach

This paper presents a delay-variation-dependent approach to fault detection of a discrete-time Markov jump neural network (MJNN) with a time-varying delay and mismatched modes. The goal is to detect the potential fault of delayed MJNNs by constructing an appropriate adaptive event-triggered and asynchronous H∞ filter. By choosing a delay-product-type Lyapunov-Krasovskii (L-K) functional with a delay-dependent matrix and exploiting some matrix polynomial inequalities, bounded real lemmas (BRLs) are obtained on the existence of suitable adaptive event generator and filters. These BRLs are dependent not only on the delay bounds but also on the delay variation rate. Simulation results are given to show the validity of the proposed theoretical method.

An algorithm for power transmission line fault detection based on improved YOLOv4 model

In response to the escalating demand for real-time and accurate fault detection in power transmission lines, this paper undertook an optimization of the existing YOLOv4 network. This involved the substitution of the main feature extraction network within the original YOLOv4 model with a lighter EfficientNet network. Additionally, the inclusion of Grouped Convolution modules in the feature pyramid structure replaced conventional convolution operations. The resulting model not only reduced model parameters but also effectively ensured detection accuracy. Moreover, in enhancing the model's reliability, data augmentation techniques were employed to bolster the robustness of the power transmission line fault detection algorithm. This optimization further utilized the DIoU loss function to stabilize target box regression. Comparative experiments demonstrated the improved YOLOv4 model's superior performance in terms of loss function optimization while significantly enhancing detection speed under equivalent configurations. The parameter capacity was reduced by 81%, totaling merely 43.65 million, while the frame rate surged by 85% to achieve 24 frames per second. These experimental findings validate the effectiveness of the algorithm.

Motor current and vibration monitoring dataset for various faults in an E-motor-driven centrifugal pump

Induction motor driven pumps are a staple in many sectors of industry, and crucial equipment in naval ships. Such machines can suffer from a wide variety of issues, which may cause it to not perform its function. This can either be due to degradation of components over time, or due to incorrect installation or usage. Unexpected failure of the machine causes downtime and lowers the availability. In some cases, it can even lead to collateral damage. To prevent collateral damage and optimise the availability, many asset owners apply condition monitoring, measuring the dynamic response of the system while in operation. Two high-frequency measurement methods are widely accepted for the detection of faults in rotating machinery at an early stage: vibration measurements, and motor current and voltage measurements. These methods can also distinguish between different failure mechanisms and severities. The dataset described in this article presents experimental data of two centrifugal pumps, driven by induction motors through a variable frequency drive. Besides measurements of behaviour that is considered healthy (new bearings, well aligned), the machines have also been subjected to a variety of (simulated) faults. These faults include bearing defects, loose foot, impeller damage, stator winding short, broken rotor bar, soft foot, misalignment, unbalance, coupling degradation, cavitation and bent shaft. Most faults were implemented at multiple levels of severity for multiple motor speeds. Both vibration measurements, and current and voltage measurements were recorded for all cases. The dataset holds value for a wide range of engineers and researchers working on the development and validation of methods for damage detection, identification and diagnostics. Due to the extensive documentation of the presented data, labelling of the data is close to perfect, which makes the data particularly suitable for developing and training machine learning and other AI algorithms.

Fault detection using PDE-based observer in transport flow

This paper deals with the state fault detection scheme for distribution flow networks subject to continuously varying conditions at boundaries. A robust PDE detection observer for transport flow systems is designed. Directly built on the nonlinear hyperbolic systems of balance laws model with anti-collocated setup, the PDE observer based on backstepping theory provide the on-line estimation of signals that are not measured. The stability of the error equation is proved. The estimation and the observability time are used for fault detection; an adaptive threshold is defined for the purpose. The performances of the observer and the fault detection method are validated on actual flow data collected from a real water distribution system (WDS) for leakage detection. The leak detection time corresponds to the first alarm activation, confirms the effectiveness of proposed approach.

Experimental investigation of a real-time singularity-based fault diagnosis method for five-phase PMSG-based tidal current applications

This paper presents a novel real-time singularity-based fault diagnosis method for tidal current applications, specifically utilizing a five-phase permanent magnet synchronous generator with trapezoidal back electromotive forces. The proposed method incorporates an innovative orthogonal signal generator through a second-order filter, enabling the extraction of detectable singularity signatures from phase current signals. The principle of the method is elucidated through step-by-step design procedures, outlining the indicator enhancement approach and adaptive thresholds employed for enhanced robustness and adaptability. Fault detection is performed based on the improved fault indicators and an adaptive threshold law, followed by immediate fault localization that is achieved via twice average operations of the phase currents. To demonstrate the effectiveness and efficiency of the proposed method, a comparative study is carried out with a classical mean current vector-based fault diagnosis method. A small-scale experimental platform emulating a tidal current application is established for a comprehensive evaluation of both methods. The experimental results highlight the superior fault diagnosis performance of the proposed method, particularly in detecting single and multiple open circuit faults in phases or switches, while exhibiting enhanced robustness against variations in torque and speed. The simplicity of implementation and rapid detection mechanism are principal merits for the proposed method.

Dynamic event-triggered fault detection for rotary steerable systems with unknown time-varying noise covariances

This paper is concerned with the fault detection problem for the rotary steerable drilling tool system under unknown vibrations and limited computational resources. Firstly, the drilling tool system can be modeled by a nonlinear stochastic system with unknown time-varying noise covariances. Then, the dynamic event-triggered mechanism is introduced to save computational resources, and the caused transmission error is completely decoupled by nonuniform sampling. Subsequently, a novel unscented Kalman filter is proposed by combining the expectation maximization method to estimate states when noise covariances are unknown. A residual and an evaluation function are constructed to detect faults. Finally, a numerical simulation and an experiment on a drilling tool prototype validate the superior performance of the proposed fault detection scheme, which has lower missed alarm rates and consumes less time than existing methods.

Fault detection for NOx emission process in thermal power plants using SIP-PCA

With the era of big data, data-driven models are increasingly vital to just-in-time decision support in pollution emission management and planning. This article aims to evaluate the usability of the proposed data-driven model to monitor NOx emission from a coal-fired boiler process using easily measured process variables. As the emission process is highly complex, process variables interact with each other, and they cannot guarantee that all the variables in the actual operation obey the Gaussian distributions. As conventional principal component analysis (PCA) can only extract variance information, a novel data-driven model is proposed, called survival information potential-based PCA (SIP-PCA) model, is proposed in this work. First, an improved PCA model is established based on the SIP performance index. SIP-PCA can extract more information in the latent space from the process variables following the non-Gaussian distributions. Then, the control limits for fault detection are determined based on the kernel density estimation method. Finally, the proposed algorithm is successfully applied to a real NOx emission process. By monitoring the operation of process variables, possible failures can be detected as soon as possible. Fault isolation and system reconstruction can be implemented in time, preventing NOx emissions from exceeding its standard.

Dataset for anomaly detection in a production wireless mesh community network

Wireless community networks, WCN, have proliferated around the world. Cheap off-the-shelf WiFi devices have enabled this new network paradigm where users build their own network infrastructure in a do-it-yourself alternative to traditional network operators. The fact that users are responsible for the administration of their own nodes makes the network very dynamic. There are frequent reboots of the networking devices, and users that join and leave the network. In addition, the unplanned deployment of the network makes it very heterogeneous, with both high and low capacity links. Therefore, anomaly detection in such dynamic scenario is challenging. In this paper we provide a dataset gathered from a production WCN. The data was obtained from a central server that collects data from the mesh nodes that build the network. In total, 63 different nodes were encountered during the data collection. The WCN is used daily to access the Internet from 17 subscribers of the local ISP available on the mesh. We have produced a dataset gathering a large set of features related not only to traffic, but other parameters such as CPU and memory. Furthermore, we provide the network topology of each sample in terms of the adjacency matrix, routing table and routing metrics. In the data we provide there is a known unprovoked gateway failure. Therefore, the dataset can be used to investigate the performance of unsupervised machine learning algorithms for fault detection in WCN. To our knowledge, this is the first dataset that allows fault detection to be investigated from a production WCN.

Intelligent fault detection scheme for constant-speed wind turbines based on improved multiscale fuzzy entropy and adaptive chaotic Aquila optimization-based support vector machine

Timely and effective fault detection is essential to ensure the safe and reliable operation of wind turbines. However, due to the complex kinematic mechanisms and harsh working environments of wind turbine equipment, it is difficult to extract sensitive features and detect faults from acquired wind turbine signals. To address this challenge, a novel intelligent fault detection scheme for constant-speed wind turbines based on refined time-shifted multiscale fuzzy entropy (RTSMFE), supervised isometric mapping (SI), and adaptive chaotic Aquila optimization-based support vector machine (ACAOSVM) is proposed. In the first step, the RTSMFE method is used to fully extract features of the wind turbine system. The time-shifted coarse-grained construction technique and a refined computing technique are adopted in the RTSMFE method to enhance the capability of traditional multiscale fuzzy entropy for measuring the complexity of signals. Subsequently, an effective manifold learning approach, SI, is applied to obtain the important and low-dimensional feature set from the high-dimensional feature set. Finally, sensitive features are fed into the ACAOSVM classifier to identify faults. The proposed ACAO algorithm is used to optimize important parameters of the SVM, thereby improving its detection performance. Simulations and wind turbine experiments verified that the proposed RTSMFE outperforms existing entropy techniques in terms of complexity measurement and feature extraction. Furthermore, the proposed ACAOSVM classifier is superior to existing advanced classifiers for fault pattern recognition. Finally, the proposed intelligent fault detection scheme can more correctly and efficiently detect wind turbine single/hybrid faults than other recently published schemes.

Bus network decomposition for fault detection and isolation through power line communication

This paper deals with fault detection and isolation (FDI) in complex embedded wired communication networks. The considered faults are soft faults which do not prevent the communication, but may evolve into hard faults, i.e. short or open circuit. A novel FDI method based on power line communication (PLC) transmission systems is proposed. In these PLC systems, the transmission coefficients between the source and each receiver are estimated for communication purposes using orthogonal frequency division multiplexing (OFDM). Health indicators and residuals are computed by comparing the online estimated transmission coefficients with the reference coefficients. A methodology for dealing with complex networks, such as bus networks, is proposed. It is based on the decomposition of the network into several Y-shaped sub-networks. Each of these sub-networks is monitored to detect the presence of a fault. The FDI method is first validated using real data extracted from a Y-shaped network test bench. Then, the proposed approach is validated on a more complex network using realistic simulated data.

Acoustic monitoring of an aircraft auxiliary power unit

In this paper, the development and implementation of a novel approach for fault detection of an aircraft auxiliary power unit (APU) has been demonstrated. The developed approach aims to target the proactive identification of faults, in order to streamline the required maintenance and maximize the aircraft's operational availability. The existing techniques rely heavily on the installation of multiple types of intrusive sensors throughout the APU and therefore present a limited potential for deployment on an actual aircraft due to space constraints, accessibility issues as well as associated development and certification requirements. To overcome these challenges, an innovative approach based on non-intrusive sensors i.e., microphones in conjunction with appropriate feature extraction, classification, and regression techniques, has been successfully demonstrated for online fault detection of an APU. The overall approach has been implemented and validated based on the experimental test data acquired from Cranfield University's Boeing 737-400 aircraft, including the quantification of sensor location sensitivities on the efficacy of the acquired models. The findings of the overall analysis suggest that the acoustic-based models can accurately enable near real-time detection of faulty conditions i.e., Inlet Guide Vane malfunction, reduced mass flows through the Load Compressor and Bleed Valve malfunction, using only two microphones installed in the periphery of the APU. This study constitutes an enabling technology for robust, cost-effective, and efficient in-situ monitoring of an aircraft APU and potentially other associated thermal systems i.e., environmental control system, fuel system, and engines.

Mixed l1/l- fault detection observer design for delayed 2D positive systems in FM LSS Models

This study is concerned with the problem of the mixed l₁/l_- fault detection (FD) observer for delayed two-dimensional (2D) positive systems (PSs). The necessary and sufficient conditions (NSCs) are derived under which the residual error system is asymptotically stable (AS) and has the prescribed performance level. The conservatism is greatly reduced compared to the existing results. A new performance analysis method and the mixed l₁/l_- FD observer design are presented. Firstly, the calculation of l₁/l_- index for delayed 2D PSs is proposed by establishing the equivalence between the delayed system and the higher dimensional delay-free system in sense of l₁ and l_- indexes. Secondly, NSCs are developed such that the delayed 2D PS is AS with a desired mixed l₁/l_- performance. Thirdly, the sufficient conditions of the observer design are further developed based on linear programming. An iterative algorithm is formulated to minimize and maximize the impact of system disturbances and faults on the output signal, respectively. Finally, we present two examples to verify the superiority of the obtained results.

Fault detection and analysis for wheelset bearings via improved explicit shift-invariant dictionary learning

The wheelset bearing is an indispensable part of the high-speed train, and monitoring its service performance is a concern of many researchers. Effective extraction of those impulse signals induced by the defects on the bearing elements is the key to fault detection and behaviour analysis. However, the presence of considerable noise and irrelevant components brings difficulties to extracting the wheelset bearing fault impulse signals from the measured vibration signals. This paper proposes an improved explicit shift-invariant dictionary learning (IE-SIDL) method to address this issue. Based on the shift-invariant characteristics of the wheelset bearing fault impulse signal in the time-domain, the circulant matrix is used to construct a shift-invariant dictionary and explicitly characterize the fault impulses at any time. To improve the efficiency of dictionary learning, a method of three flips is introduced to realize fast dictionary construction, and the frequency-domain reconstruction property of the circulant matrix is employed to quickly update the dictionary. Besides, an indicator-guided subspace pursuit (SP) method based on the sparsity of envelope spectrum (SES) is adopted for the sparse coding to improve sparse solution accuracy and adaptation. The effectiveness of the IE-SIDL method is proved through the simulated and experimental signals. The results demonstrate that the improved dictionary learning method has an excellent capacity in extracting fault impulse signal of the wheelset bearings, and the good time- and frequency-domain characteristics of the processed signals facilitate fault detection and behaviour analysis.

Asynchronous fault detection filtering for Markovian jump systems with output sensor saturation

In this paper, a robust asynchronous filter is proposed to detect faults in Markovian jump systems with sensor saturations. The proposed filter has a mode different from that of the original system, and is tolerant of external disturbances. The residual is generated to construct an evaluation function that can successfully perform fault detection. Detailed filter matrices can be computed from the existing conditions. Furthermore, practical and comparative simulations are performed to verify the efficiency of the proposed technique.

Actuator failure compensation-based command filtered control of electro-hydraulic system with position constraint

In this article, the design and experimental evaluation of a fault-tolerant controller are introduced for a double-rod electro-hydraulic actuator subjected to actuator faults and disturbances. The internal leakage fault is captured as a bias fault, whilst the faults in servo-valve and supply failure are considered as a partial loss of effectiveness (LOE) fault. The design obstacles caused by the disturbances and bias fault are suppressed by nonlinear disturbance observers (NDO) while an asymmetric barrier Lyapunov function is used to ensure the non-violated boundary of the output position. To tackle the LOE fault, the development of an enhanced adaptive compensation technique for actuator fault-tolerant control (FTC) is then constructed. Moreover, to mitigate the "explosion of complexity" in the traditional backstepping design, the command-filtered control is utilized to elaborate the FTC scheme. It is shown by theoretical analysis that system stability is ensured under faulty conditions. Finally, simulation/experiment results and comparison studies are performed to further verify the effectiveness of the proposed approach.

The use of artificial neural network for low latency of fault detection and localisation in transmission line

One of the most critical concerns in power system reliability is the timely and accurate detection of transmission line faults. Therefore, accurate detection and localisation of these faults are necessary to avert system collapse. This paper focuses on using Artificial Neural Networks in faults detection and localisation to attain accuracy, precision and speed of execution. A 330 kV, 500 km three-phase transmission line was modelled to extract faulty current and voltage data from the line. The Artificial Neural Network technique was used to train this data, and an accuracy of 100% was attained for fault detection and about 99.5% for fault localisation at different distances with 0.0017 μs of detection and an average error of 0%-0.5%. This model performs better than Support Vector Machine and Principal Component Analysis with a higher fault detection time. This proposed model serves as the basis for transmission line fault protection and management system.

Selective weighted multi-scale morphological filter for fault feature extraction of rolling bearings

Morphological filtering shows effectiveness in vibration signal analysis because of its simplicity and efficiency. Considering that different structural elements have different effects on filtering results, a new multi-scale morphological filtering (MMF) method called selective weighted multi-scale morphological filter (SWMMF) is developed for integrating results of different scales based on adaptive weighting strategy. Firstly, four morphological operators (dilation-closing, closing-dilation, erosion-opening and opening-erosion) are integrated into a new combination difference morphological filter to strengthen effect of faulty component extraction. Secondly, this new morphological filter is further extended to multiple scales in order to overcome limitation of single scale filter. Finally, the filtered results of different scales are adaptively combined by using the whale optimization algorithm (WOA)-based selective weighting method. The effectiveness of multi-scale filter and selective weights is proved by comparing with single-scale and average weighting filter on simulation and real-world cases (bearing vibration signals with different defects). The testing results on vibration signals indicate that SWMMF is able to extract effectively defect frequency and the corresponding multiplication frequencies from bearing vibration signals with heavy noise. The testing results illustrate that SWMMF outperforms other representative MMFs (e.g., weighted multi-scale morphological gradient operator (WMMG), weighted multi-scale difference operator (WMDIF), weighted multi-scale average operator (WMAVG)) on impulsive feature extraction of bearing vibrations signals with various defects. Moreover, it is demonstrated that SWMMF has good applicability in bearing fault diagnosis due to setup of adaptive weights and selection of structure element.

A latent feature oriented dictionary learning method for closed-loop process monitoring

Industrial cyber-physical system (ICPS), by its powerful computing, communication, precise control and remote operation functions, has become the mainstream of modern industrial process. The observed variables of the closed-loop process in ICPS are subject to the degradation of equipment and other factors, resulting in exhibiting a stationary/nonstationary mixture feature and dynamic feature. Moreover, due to the frequent change of working conditions in the closed-loop process, the traditional open-loop process monitoring method always triggers false alarms, which will impose a negative impact on the safety and trustworthiness of ICPS. Therefore, for the closed-loop process in ICPS, a latent feature oriented dictionary learning (LFDL) method is proposed, which realizes the precise separation of latent features of raw data through three stages. First, closed-loop process variables are separated into stationary and nonstationary variables to mine the local information spatially. Then, from the temporal viewpoint, the static and dynamic features were extracted for stationary and nonstationary variables on the basis of the slow feature analysis method and cointegration analysis for local monitoring. Finally, the global monitoring results are obtained by utilizing the dictionary learning method to fuse respectively the local monitoring results of the static and dynamic features. Since the proposed method has taken the feature of the close-loop process from temporal and spatial viewpoints simultaneously, it can distinguish the normal change of operating conditions and actual faults accurately. Extensive experiments including the three-phase flow, the Tennessee Eastman process and an industrial roasting process are conducted to demonstrate the feasibility and effectiveness of the proposed method.

Detection of faults and DG islanding in PV-Wind DC ring bus microgrid by using optimized VMD based improved broad learning system

This paper presents a novel approach for the detection and classification of photovoltaic with wind based DC ring bus microgrid DC faults and DG (distributed generation) islanding events. This novel approach consists of adaptive variational mode decomposition (AVMD) and an improved broad learning system (IBLS). Initially, DC fault current signals are captured from the DC bus under different operating conditions and processed through the AVMD to decompose the signals into intrinsic mode functions (IMFs). The VMD is made adaptive by minimizing the objective function of the L-kurtosis index for optimal modal number (K) and penalty factor (σ) through the improved whale optimization (IWO) algorithm. From the optimal IMFs, the most significant IMFs are chosen based on the threshold of the L-kurtosis index, and they are passed through statistical features to extract efficient data. Further, the training and testing of this data set is carried out through IBLS for obtaining the accurate detection and discrimination of DC faults. The conventional BLS method is improved through elastic net ridge regression for calculating the weights. The effectiveness of the proposed AVMD based IBLS algorithm is verified by its superiority in terms of relative computation time (RCT), classification accuracy (CA) with the confusion matrix, and their performance indices by comparing with other existing methods under different case studies. Finally, the simplicity and practicability of the proposed work are tested and implemented in the dSPACE 1104 embedded processor.

Optimally detecting and classifying the transmission line fault in power system using hybrid technique

In this paper, a hybrid system is proposed to predict and classifies the power system transmission line faults. The proposed technique is the consolidation of both the truncated singular value decomposition (TSVD) and Human urbanization algorithm (HUA) based Recurrent Perceptron Neural Network (RPNN), and hence it is named as TSVD-HUARPNN technique. TSVD is matrix decomposition, this technique qualify the outcome it as fast or not. In the proposed work, the qualification of the results from the TSVD is improved by a lemma theorem; it is a proven proposition which is used to obtain a larger and optimal result. For that reason, it is also known as a "helping theorem" or an "auxiliary theorem". Here, it has two modules for power system fault analysis: (i) fault detection, (ii) fault classification. The first process of the proposed system is the generation of the dataset of normal and abnormal conditions of transmission line parameters of power system using TSVD. The extracted dataset is assessed by HUA-based RPNN system to classify the fault analysis that occurs in transmission system. The TSVD-HUARPNN system is used to predict and classify the fault present in the transmission line. The proposed TSVD-HUARPNN system ensures the system with less complexity for the detection and classification of the fault, therefore the accuracy of the system is increased. By then, the proposed model is activated in MATLAB/Simulink, its performance is evaluated with the existing models. The performance with noise at 20 dB of the proposed technique is 99.77%.

H-/H∞ fault detection observer design for fractional-order singular systems in finite frequency domains

This paper investigates the robust fault detection observer design problem in finite frequency domains for fractional-order singular systems. An H_-/H_∞ fault detection observer for fractional-order singular systems is designed to meet the system admissibility, disturbance robustness and fault sensitivity indices in finite frequency domains simultaneously. Four theorems and four corollaries about the system admissibility and the performance index H_-/H_∞ in different frequency ranges are given, and then the sufficient conditions are obtained in terms of linear matrix inequalities. Finally, two numerical examples are given to verify the validity of the presented methods.

Fault detection for chemical processes based on non-stationarity sensitive cointegration analysis

Due to the time-varying operation conditions, chemical processes are characterized by non-stationary characteristics, which makes it a great challenge for conventional process monitoring methods to capture the non-stationary variations In the non-stationary processes, the abnormality would cause the stationary variables to be non-stationary. In this article, a non-stationarity sensitive cointegration analysis monitoring method is proposed to explore potential non-stationary variations. First, the essential non-stationary variables are distinguished using Augmented Dickey-Fuller test to eliminate the influence of essential non-stationary under normal conditions. Then by further analyzing the faulty data, the variables which are sensitive to the non-stationary variations are selected. On this basis, cointegration analysis models are established for both the essential non-stationary variables and non-stationarity sensitive variables to explore long-term dynamic equilibrium relationship, respectively. With the selection of non-stationarity sensitive variables, the potential faulty information is emphasized in the process monitoring model, which makes the model capable to handle the non-stationary variations. Finally, the monitoring results are combined through Bayesian inference criterion. The proposed method is applied on the Tennessee Eastman process and a vinyl acetate monomer plant model, and the feasibility and performance are demonstrated.

Oscillatory Failure Case detection in flight control systems via wavelets decomposition

A fault detection design is proposed for addressing the Oscillatory Failure Case (OFC) detection problem, introduced in the joint Airbus-Stellenbosch university aerospace industrial-benchmark competition called at the IFAC 2020 World Congress¹. The detection scheme is comprised of an output estimator, a wavelet decomposition and an energy-based denoising method, and the residual evaluation unit. The detection problem of wide frequency range OFCs is also addressed. According to the achieved simulation results, the proposed fault detection method is able to satisfy the competition prescriptions in the frequency range [1 10] Hz for those OFC's having an amplitude greater than 2.3 mm for OFCs at rod position sensor, or 1.4 mA for OFCs at servo input current, regardless of disturbances level, uncertainties and load factor control input. In other cases, faults are detected slightly after the prescribed detection limit, with some interesting exceptions.

Design of a computationally efficient observer-based distributed fault detection and isolation scheme in second-order networked control systems

This paper presents a novel directional observer-based fault detection and isolation scheme for second-order networked control systems (NCS). The directional unknown input observer (UIO) tool is exploited to study the problem of distributed fault detection and isolation (FDI). Two design schemes with global and partial/local network models are proposed to solve the distributed FDI problem. Thresholds are computed for the application of the proposed schemes in a noisy environment. In addition, the salient features of the proposed schemes are that both fault detection and fault isolation are achieved in a single step using a single observer. The schemes are applied to power system models to validate their results. A detailed comparison with existing FDI schemes is also provided, which clearly shows the effectiveness of the proposed scheme in terms of computational requirements.

Criteria for optimizing kernel methods in fault monitoring process: A survey

Nowadays, how to select the kernel function and their parameters for ensuring high-performance indicators in fault diagnosis applications remains as two open research issues. This paper provides a comprehensive literature survey of kernel-preprocessing methods in condition monitoring tasks, with emphasis on the procedures for selecting their parameters. Accordingly, twenty kernel optimization criteria and sixteen kernel functions are analyzed. A kernel evaluation framework is further provided for helping in the selection and adjustment of kernel functions. The proposal is validated via a KPCA-based monitoring scheme and two well-known benchmark processes.

Fault detection for linear discrete time-varying systems with multiplicative noise based on parity space method

This paper addresses the robust fault diagnosis problem for a class of linear discrete time-varying systems with multiplicative noise based on parity space method. A novel fault detection performance index, in terms of stochastic robustness/sensitivity ratio, is proposed to establish the residual generator. A computationally attractive recursive algorithm, is put forward to obtain the complex matrix involved in the aforementioned fault detection performance index. Drawing support of random matrix analysis and calculation, the corresponding solution is derived in an analytical form via solving a multi-objective optimization problem. By means of Randomized Algorithms, two fault detection threshold setting algorithms are provided subsequently to achieve residual performance assessment by taking into account the fault detection rate and false alarm rate in the probabilistic framework. Two illustrative examples are finally provided to illustrate the effectiveness of the proposed scheme.

High-impedance faults in power distribution systems: A narrative of the field's developments

High-impedance faults in power distribution systems is a lasting problem with decades of steady investigation. Due to the complexity of the problem, the field can also be challenging to navigate. Although there exist surveys of the field in the literature, it is not easy to find a comprehensive contextualization of how and when the field developments unfolded. This paper presents the historical narrative of the progress and developments based on the most cited papers since the inception of the field. The accounts are not limited to archaic and obsolete works. They are all contextualized from the seminal papers to contemporary methods and related technology. Quantitative figures on the survey of the methods and relevant knowledge gaps are also discussed at the closing of the paper.

Robust adaptive boosted canonical correlation analysis for quality-relevant process monitoring of wastewater treatment

Quality-relevant process monitoring has attracted much attention for its ability to assist in maintaining efficient plant operation. However, when the process suffers from non-stationary and over-complex (with noise, multiplicative faults, etc.) characteristics, the traditional methods usually cannot be effectively applied. To this end, a novel method, termed as Robust adaptive boosted canonical correlation analysis (Rab-CCA), is proposed to monitor the wastewater treatment processes. First, a robust decomposition method is proposed to mitigate the defects of standard CCA by decomposing the corrupted matrix into a low-matrix and a sparse matrix. Second, to further improve the performance of the standard process monitoring method, a novel criterion function and control charts are reconstructed accordingly. Moreover, an adaptive statistical control limit is proposed that can adjust the thresholds according to the state of a system and can effectively reduce the missed alarms and false alarms simultaneously. The superiority of Rab-CCA is verified by Benchmark Simulation Model 1 (BSM1) and a real full-scale wastewater treatment plant (WWTP).

Nonlinear quality-related fault detection using combined deep variational information bottleneck and variational autoencoder

Deep learning has gotten much attention in industrial field, many fault detection methods based on deep learning have been developed for nonlinear industrial processes. However, most of them do not take the quality-related faults into account. In order to extract the latent variables which can represent the separated quality-related and unrelated information, this paper proposes a novel deep VIB-VAE algorithm, which combines variational autoencoder (VAE) model and deep variational information bottleneck (VIB). Deep VIB extracts quality-related latent variables by maximizing mutual information between latent variables and process quality while minimizing mutual information between latent variables and observation. VAE is used to learn the quality-unrelated part with above quality-related latent variables as auxiliary information. To monitor and distinguish quality-related and quality-unrelated faults, two monitoring statistics are designed by the two-part latent variables. The reconstruction error by VAE is introduced to improve the performance of fault detection. Finally, the effectiveness of the proposed deep VIB-VAE algorithm is demonstrated by a numerical case and a real hot strip mill process case, respectively.

Neural network-based fault detection for nonlinear networked systems with uncertain medium access constraint: Application to motor systems

The fault detection for a class of continuous-time nonlinear networked control systems with medium access constraint is concerned in this paper, where the occurring probability of transition from one sensor to another is allowed to be partially unknown and uncertain. First of all, a Markovian system approach is adopted to describe the access process of sensors, in which only one sensor is allowed to access the communication channel. A robust filter based residual generator is proposed to generate the residual signal such that it can be used to indicate whether the fault has occurred or not. The nonlinear term is approximated by a neural network, and the Lyapunov-Krasovskii functional is introduced to analyze the fault detection system, three sufficient conditions for the stochastic stability of fault detection error system are given, and the fault detection filter gains are calculated via solving some matrix inequalities. In the simulation, a second-order DC motor system is used to validate of the main results, which shows the effectiveness of the fault detector design.

From fault detection to one-class severity discrimination of 3D printers with one-class support vector machine

The lack of faulty condition data reduces the feasibility of supervised learning for fault detection or fault severity discrimination in new manufacturing technologies. To deal with this issue, one-class learning arises for building binary discriminative models using only healthy condition data. However, these models have not been extrapolated to severity discrimination. This paper proposes to extend OCSVM, which is typically used for fault detection, to 3D printer fault severity discrimination. First, a set of features is extracted from a set of normal signals. An optimized OCSVM model is obtained by tuning the kernel and model hyperparameters. The resulting models are evaluated for fault detection and fault severity discrimination using a proposed performance evaluation approach. Experimental comparisons for belt-based faults in 3D printers show that the distance to the hyperplane has the information to discriminate the severity level, and its use is feasible. The proposed hyperparameter optimization technique improves the OCSVM for fault detection and severity discrimination compared to some other methods.

An effective method for detection of stator fault in PMSM with 1D-LBP

Permanent Magnet Synchronous Motors (PMSMs) have recently been used commonly in all areas of the industry due to their position control as well as precise speed. The success of these motors in applications of precise speed and position control depends on their whole operation. Even if the fault is at a highly-low-level, this negatively affects the precision of the motor. In this study, the one dimensional local binary patterns (1D-LBP) method, which is compelling and distinctive, has been used for feature extraction instead of frequency spectrum analysis or time-frequency analysis, which are among conventional feature extraction techniques in the literature, to detect short-circuit fault that occurs in PMSM stators. Thus, to test the proposed method, an experiment setup has been prepared to record current and voltage signals detected through 15 kHz sampling from healthy and faulty PMSM. 1D-LBP was applied to these current and voltage signals and the histograms of newly formed current and voltage signals were obtained. Histograms of newly formed signals are used as feature vectors. Healthy and faulty motors could be classified at high success rates applying one of the machine learning techniques, Knn algorithm, to histograms. It was found that the methods had a success rate over 90% when it was tested over-current and voltage data obtained from PMSM that ran at different speeds and loads and had different fault rates to test whether the methods ran properly.

Distributed fault detection and estimation in cyber-physical systems subject to actuator faults

The fault detection and estimation problems for the physical layer network in the cyber-physical systems with unknown external disturbances are investigated in this study. Both bias fault and loss of efficiency scenarios are considered for the actuators. Based on the adaptive threshold method and sliding mode observer approach, a distributed fault detection observer (DFDO) is constructed for each physical layer node to detect the occurrence of actuator faults. Then a relative global estimation error system is defined for the distributed fault estimation observer (DFEO). Compared with the existing results, the proposed DFEO can provide the estimation for not only the actuator bias faults but also the actuators' efficiency factors under the impact of exogenous disturbance with two gain dynamic update processes. Finally, the feasibility and effectiveness of the given DFDO and the DFEO are examined by Lyapunov stability method and the simulation results.

Distributed fault detection and isolation in second order networked systems in a cyber-physical environment

Modern industrial processes and cyber-physical systems (CPS) are prone to anomalies both due to cyber and physical perturbations. Cyber disturbances or attacks being more hazardous may give birth to a series of multiple coordinated faults. In order to detect and isolate such faults, this paper proposes a novel distributed fault detection and isolation scheme for second-order networked systems. The system is assumed to be working in a cyber-physical environment where it is likely to face multiple simultaneous faults. Each node has access to measurements of states of its neighboring nodes. A distributed fault detection and isolation filter (DFDIF) is designed such that fault detection and fault isolation can be obtained in a single step. Using the proposed filter, each node can detect and isolate multiple simultaneous faults in its neighboring nodes. The detection and isolation of faults with a single filter at each node reduces the overall computational burden of distributed fault detection and isolation (DFDI) scheme. The proposed framework is tested for power network and robotic formations. Finally, a comparison with existing techniques is provided to prove the effectiveness of the proposed method.

Batch process fault detection for multi-stage broad learning system

In the real industrial production process, some minor faults are difficult to be detected by multivariate statistical analysis methods with mean and variance as detection indicators due to the aging equipment and catalyst deactivation. With structural characteristics, deep neural networks can better extract data features to detect such faults. However, most deep learning models contain a large number of connection parameters between layers, which causes the training time-consuming and thus makes it difficult to achieve a fast-online response. The Broad Learning System (BLS) network structure is expanded without a retraining process and thus saves a lot of training time. Considering that different stages of the batch production process have different production characteristics, we use the Affinity Propagation (AP) algorithm to separate the different stages of the production process. This paper conducts research on a multi-stage process monitoring framework that integrates AP and the BLS. Compared with other monitoring models, the monitoring results in the penicillin fermentation process have verified the superiority of the AP-BLS model.

Fault detection for planetary gearbox based on an enhanced average filter and modulation signal bispectrum analysis

Transient impulses are important information for machinery fault diagnosis. However, the transient features contained in the vibration signals generated by planetary gearboxes are usually immersed by a large amount of background noise and harmonic components. Even mathematical morphology (MM) is an excellent anti-noise signal processing method that can directly extract the geometry of impulse features in the time domain, but the four basic operators of MM can only extract one-way impulses while cannot extract the bidirectional impulses effectively at the same time. To accurately extract the impulse feature information, a novel method for fault detection of planetary gearbox based on an enhanced average (EAVG) filter and modulated signal bispectrum (MSB) is proposed. Firstly, the properties of the extracted impulses based on the four basic operators of MM will be divided into two categories of enhanced average operators. The four EAVG filters consist of the average weighted combination of enhanced average operators, and then the best EAVG filter is selected based on correlation coefficient to implement on the original vibration signal. It allows EAVG filter to extract positive and negative impulses of vibration signal, thereby improving the accuracy of planetary gearbox fault detection. Subsequently, the performance of the EAVG filter is influenced by the length of its structural element (SE), which is adaptively determined using an indicator based kurtosis. Then, the EAVG filter selects the optimal SE length to eliminate the interference of background noise and harmonic components to enhance the impulse components of the vibration signal. However, the nonlinear modulation components that are related to the fault types and severities are not extracted exactly and still remained in the filtered signal by EAVG. Finally, the MSB is utilized to the EAVG filtered signal to decompose the modulated components and extract the fault features. The advantages of EAVG over average (AVG) filter are clarified in the simulation study. In addition, the EAVG-MSB is validated by analyzing the vibration signals of planetary gearboxes with sun gear chipped tooth, sun gear misalignment and bearing inner race fault. The results indicate that the EAVG-MSB is effective and accurate in feature extraction compared with the combination morphological filter-hat transform (CMFH) and average combination difference morphological filter (ACDIF), and the feasibility of the EAVG-MSB are proved for planetary gearbox condition monitoring and fault diagnosis.

Periodical sparse low-rank matrix estimation algorithm for fault detection of rolling bearings

Early bearing fault detection is crucial to avoid catastrophic accidents. However, the repetitive defect impulses indicating bearing fault are buried in heavy background noise. In the paper, a novel periodical sparse low-rank (PSLR) matrix estimation algorithm is proposed for extracting repetitive transients from noisy signal. Concretely, periodical group sparsity and low-lank property of fault transients in time-frequency domain are first revealed, and then an optimization problem is proposed for simultaneously promoting these two properties. Meanwhile, to further highlight the sparsity of fault features, the non-convex penalty functions are incorporated into the optimization problem. Then, for solving the proposed optimization problem, an iterative algorithm is derived based on alternating direction method of multipliers (ADMM) and majorization-minimization (MM), in which the traditional soft-thresholding operation is replaced by the proposed Gini-guided fault information thresholding (FIT) scheme to enhance fault transient extraction. Finally, simulated and real signals confirm the performance of proposed PSLR in extracting defect impulses from noisy vibration signal.

Rolling element bearing fault identification using a novel three-step adaptive and automated filtration scheme based on Gini index

For early detection of rolling element bearings (REBs) faults in contaminated signals, kurtosis-derived indices are involved in the filtration process prior to demodulation. However, they were found either sensitive to impulsive outliers or requiring many input arguments. In this study, a novel three-step adaptive and automated filtration scheme using Gini index (GI) is proposed as an alternative to kurtosis-based techniques to enhance the weak fault features and eliminate noise and interferences from the raw vibration signal. The proposed approach was tested using experimental signals with different bearing faults. The filtered signals were greatly denoised and the fault impulses were successfully isolated, which indicates the effectiveness of the proposed approach and the superiority of GI over kurtosis-derived indices as a criterion for proper filter design for REBs fault detection.

DCNN for condition monitoring and fault detection in rotating machines and its contribution to the understanding of machine nature

Rotating machines are critical equipment in many processes, and failures in their operation can have serious implications. Consequently, fault detection in rotating machines has been widely investigated. Conventional detection systems include two blocks: feature extraction and classification. These systems are based on manually engineered features (ball pass frequencies, RMS value, kurtosis, crest factor, etc.) and therefore require a high level of human expertise (it is a human who designs and selects the most appropriate set of features to perform the classification). Instead, we propose a system for condition monitoring and fault detection in rotating machines based on a 1-D deep convolutional neural network (1D DCNN), which merges the tasks of feature extraction and classification into a single learning body. The proposed system has been designed for use on a rotating machine with seven possible operating states and it proves to be able to determine the operating condition of the machine almost as accurately as conventional feature-engineered classifiers, but without the need for prior knowledge of the machine. The proposed system has also reported good classification on a bearing fault dataset from another machine, thus demonstrating its capability to monitor the condition of different machines. Finally, the analysis of the features learned by the deep model has revealed valuable and previously unknown machine information, such as the rotational speed of the machine or the number of balls in the bearings. In this way, our results illustrate not only the good performance of CNNs, but also their versatility and the valuable information they could provide about the monitored machine.

Multivariate statistical monitoring of subway indoor air quality using dynamic concurrent partial least squares

To maintain the health level of indoor air quality (IAQ) in subway stations, the data-driven multivariate statistical method concurrent partial least squares (CPLS) has been successfully applied for output-relevant and input-relevant sensor faults detection. To cope with the dynamic problem of IAQ data, the augmented matrices are applied to CPLS (DCPLS) to achieve the better performance. DCPLS method simultaneously decomposes the input and output data spaces into five subspaces for comprehensive monitoring: a joint input-output subspace, an output principal subspace, an output-residual subspace, an input-principal subspace, and an input-residual subspace. Results of using the underground IAQ data in a subway station demonstrate that the monitoring capability of DCPLS is superior than those of PLS and CPLS. More specifically, the fault detection rates of the bias of PM₁₀ and PM_2.5 using DCPLS can be improved by approximately 13% and 15%, respectively, in comparison with those of CPLS.

Monitoring and detecting faults in wastewater treatment plants using deep learning

Wastewater treatment plants use many sensors to control energy consumption and discharge quality. These sensors produce a vast amount of data which can be efficiently monitored by automatic systems. Consequently, several different statistical and learning methods are proposed in the literature which can automatically detect faults. While these methods have shown promising results, the nonlinear dynamics and complex interactions of the variables in wastewater data necessitate more powerful methods with higher learning capacities. In response, this study focusses on modelling faults in the oxidation and nitrification process. Specifically, this study investigates a method based on deep neural networks (specifically, long short-term memory) compared with statistical and traditional machine-learning methods. The network is specifically designed to capture temporal behaviour of sensor data. The proposed method is evaluated on a real-life dataset containing over 5.1 million sensor data points. The method achieved a fault detection rate (recall) of over 92%, thus outperforming traditional methods and enabling timely detection of collective faults.

Active fault diagnosis for uncertain systems using optimal test designs and detection through classification

Fault detection and isolation (FDI) is becoming increasingly difficult due to the complexity and uncertainty of modern systems. For industrial systems with explicit models available, model-based active FDI tests can improve the capability for fault diagnosis. These tests should be determined and evaluated prior to implementation to minimize on-site computational costs. In this paper, a methodology is presented for the design optimization and assessment of tests for active fault diagnosis. The objective is to maximize the information from system outputs with respect to faults while minimizing the correlation between faults and uncertainty. After a test is designed, it is deployed with a k-nearest neighbor algorithm combined with principal component analysis.Two case studies are used to verify the proposed methodology, a three-tank system and a diesel engine.

Fault detection for non-Gaussian stochastic distribution fuzzy systems by an event-triggered mechanism

This paper studies an event-triggered fault detection (FD) problem for non-Gaussian stochastic distribution fuzzy systems. Different from other systems, the available information of the stochastic distribution systems is the measurable output probability density functions (PDFs) rather than the output itself. This increases the difficulty of the event-triggered-based observer synthesis. To overcome the difficulty, a new event-triggered observer approach based on the information of the output PDFs is proposed. First, a B-spline model is employed to approximate the output PDFs. Second, a novel event-triggered scheme (ETS) is designed to save the limited communication source. Then, a finite-frequency H_∕L_∞ fault detection observer is constructed such that the effect of the PDFs approximation error on the residual signal can be attenuated and the FD performance can be increased. Finally, two examples are presented to demonstrate the effectiveness of the proposed method.

Online sensor validation in sensor networks for bioprocess monitoring using swarm intelligence

Sensor faults can impede the functionality of monitoring and control systems for bioprocesses. Hence, suitable systems need to be developed to validate the sensor readings prior to their use in monitoring and control systems. This study presents a novel approach for online validation of sensor readings. The basic idea is to compare the original sensor reading with predictions for this sensor reading based on the remaining sensor network's information. Deviations between original and predicted sensor readings are used to indicate sensor faults. Since especially batch processes show varying lengths and different phases (e.g., lag and exponential phase), prediction models that are dependent on process time are necessary. The binary particle swarm optimization algorithm is used to select the best prediction models for each time step. A regularization approach is utilized to avoid overfitting. Models with high complexity and prediction errors are penalized, resulting in optimal predictions for the sensor reading at each time step (5% mean relative prediction error). The sensor reliability is calculated by the Kullback-Leibler divergence between the distribution of model-based predictions and the distribution of a moving window of original sensor readings (moving window size = 10 readings). The developed system allows for the online detection of sensor faults. This is especially important when sensor data are used as input to soft sensors for critical quality attributes or the process control system. The proof-of-concept is exemplarily shown for a turbidity sensor that is used to monitor a Pichia pastoris-batch process.

A combined mono- and multi-turbine approach for fault indicator synthesis and wind turbine monitoring using SCADA data

The monitoring of wind turbines using SCADA data has received lately a growing interest from the fault diagnosis community because of the very low cost of these data, which are available in number without the need for any additional sensor. Yet, these data are highly variable due to the turbine constantly changing its operating conditions and to the rapid fluctuations of the environmental conditions (wind speed and direction, air density, turbulence, …). This makes the occurrence of a fault difficult to detect. To address this problem, we propose a multi-level (turbine and farm level) strategy combining a mono- and a multi-turbine approach to create fault indicators insensitive to both operating and environmental conditions. At the turbine level, mono-turbine residuals (i.e. a difference between an actual monitored value and the predicted one) obtained with a normal behavior model expressing the causal relations between variables from the same single turbine and learnt during a normal condition period are calculated for each turbine, so as to get rid of the influence of the operating conditions. At the farm level, the residuals are then compared to a wind farm reference in a multi-turbine approach to obtain fault indicators insensitive to environmental conditions. Indicators for the objective performance evaluation are also proposed to compare wind turbine fault detection methods, which aim at evaluating the cost/benefit of the methods from a production manager's point of view. The performance of the proposed combined mono- and multi-turbine method is evaluated and compared to more classical methods proposed in the literature on a large real data set made of SCADA data recorded on a French wind farm during four years : it is shown than it can improve the fault detection performance when compared to a residual analysis limited at the turbine level only.

Fault detection in dynamic plant-wide process by multi-block slow feature analysis and support vector data description

This study describes a dynamic large-scale process fault detection algorithm based on multi-block slow feature analysis by taking advantages of both multi-block algorithms in highlighting the local information and slow feature analysis in extracting the different dynamics of process data. A mutual information-based relevance matrix is first calculated to measure the correlation between any two variables, and then K-means clustering is used to cluster the original variables into several blocks by gathering the variables with similar relevance vectors into the same block. Slow feature analysis is applied in each block. A support vector data description is utilized to give a final decision. The proposed algorithm is tested with a well-known Tennessee Eastman (TE) process. The fault detection results show the efficiency and the superiority of the proposed method as compared to other related methods.

Fault detection observer design for 2-D continuous nonlinear systems with finite frequency specifications

This research studies the problem of fault detection observer design for two-dimensional (2-D) continuous-time nonlinear systems in Takagi-Sugeno (T-S) form. Finite frequency (FF) specifications are used to design the observers, which makes observer designing different from previously proposed 2-D detection observers. Faults and disturbances are considered to be dominated in two different FF domain intervals. Fault sensitivity and disturbance robustness are measured by two FF performance indices, respectively. The aim of this paper is to design fault detection observers such that the residual error system has the sensitivity to faults and the desired robustness to disturbances. Sufficient conditions for the existence of a desired fault detection fuzzy observer are established in terms of linear matrix inequalities (LMIs). Simulation results indicate that faults can be detected effectively using the proposed method.

Neighborhood preserving neural network for fault detection

A novel statistical feature extraction method, called the neighborhood preserving neural network (NPNN), is proposed in this paper. NPNN can be viewed as a nonlinear data-driven fault detection technique through preserving the local geometrical structure of normal process data. The "local geometrical structure " means that each sample can be constructed as a linear combination of its neighbors. NPNN is characterized by adaptively training a nonlinear neural network which takes the local geometrical structure of the data into consideration. Moreover, in order to extract uncorrelated and faithful features, NPNN adopts orthogonal constraints in the objective function. Through backpropagation and eigen decomposition (ED) technique, NPNN is optimized to extract low-dimensional features from original high-dimensional process data. After nonlinear feature extraction, Hotelling T² statistic and the squared prediction error (SPE) statistic are utilized for the fault detection tasks. The advantages of the proposed NPNN method are demonstrated by both theoretical analysis and case studies on the Tennessee Eastman (TE) benchmark process. Extensive experimental results show the superiority of NPNN in terms of missed detection rate (MDR) and false alarm rate (FAR). The source code of NPNN can be found in https://github.com/htzhaoecust/npnn.