Hidden Markov Model-Based Fault Detection Approach for a Multimode Process

A Monitoring Method Based on FDALM and Its Application in the Sintering Process of Ternary Cathode Material

In industrial processes, the composition of raw material and the production environment are complex and changeable, which makes the production process have multiple steady states. In this situation, it is difficult for the traditional single-mode monitoring methods to accurately detect the process abnormalities. To this end, a multimode monitoring method based on the factor dynamic autoregressive hidden variable model (FDALM) for industrial processes is proposed in this paper. First, an improved affine propagation clustering algorithm to learn the model modal factors is adopted, and the FDALM is constructed by combining multiple high-order hidden state Markov chains through the factor modeling technology. Secondly, a fusion algorithm based on Bayesian filtering, smoothing, and expectation-maximization is adopted to identify model parameters. The Lagrange multiplier formula is additionally constructed to update the factor coefficients by using the factor constraints in the solving. Moreover, the online Bayesian inference is adopted to fuse the information of different factor modes and obtain the fault posterior probability, which can improve the overall monitoring effect of the model. Finally, the proposed method is applied in the sintering process of ternary cathode material. The results show that the fault detection rate and false alarm rate of this method are improved obviously compared with the traditional methods.

Linear Chain Conditional Random Field for Operating Mode Identification and Multimode Process Monitoring

As a supervised machine learning algorithm, conditional random fields are mainly used for fault classification, which cannot detect new unknown faults. In addition, faulty variable location based on them has not been studied. In this paper, conditional random fields with a linear chain structure are utilized for modeling multimode processes with transitions. A linear chain conditional random field model is trained by normal data with mode label. This model is able to distinguish transitions from stable modes well. After mode identification, the expectation of state feature function is developed for fault detection and faulty variable location. Case studies on the Tennessee Eastman process and continuous stirred tank reactor (CSTR) testify the effectiveness of the proposed approach.

One-dimensional convolutional neural network-based active feature extraction for fault detection and diagnosis of industrial processes and its understanding via visualization

Feature extraction from process signals enables process monitoring models to be effective in industrial processes. Deep learning presents extensive possibilities for extracting abstract features from image and visual data. However, the main inputs of conventional deep neural networks are large images. To overcome this, a one-dimension convolution neural network-based model optimized by a reinforcement-learning-based neural architecture search, is proposed for multivariate processes control. The experimental results illustrate its predominance for detecting and recognizing process faults. Feature and network visualization are also implemented to explore the reasons for its outstanding performance. This research extends the applications of convolutional neural network based on one-dimension process signals in complex multivariate process control.

A Review on Data-Driven Process Monitoring Methods: Characterization and Mining of Industrial Data

Safe and stable operation plays an important role in the chemical industry. Fault detection and diagnosis (FDD) make it possible to identify abnormal process deviations early and assist operators in taking proper action against fault propagation. After decades of development, data-driven process monitoring technologies have gradually attracted attention from process industries. Although many promising FDD methods have been proposed from both academia and industry, challenges remain due to the complex characteristics of industrial data. In this work, classical and recent research on data-driven process monitoring methods is reviewed from the perspective of characterizing and mining industrial data. The implementation framework of data-driven process monitoring methods is first introduced. State of art of process monitoring methods corresponding to common industrial data characteristics are then reviewed. Finally, the challenges and possible solutions for actual industrial applications are discussed.

Maintenance Prediction through Sensing Using Hidden Markov Models—A Case Study

The availability maximization is a goal for any organization because the equipment downtime implies high non-production costs and, additionally, the abnormal stopping and restarting usually imply loss of product’s quality. In this way, a method for predicting the equipment’s health state is vital to maintain the production flow as well as to plan maintenance intervention strategies. This paper presents a maintenance prediction approach based on sensing data managed by Hidden Markov Models (HMM). To do so, a diagnosis of drying presses in a pulp industry is used as case study, which is done based on data collected every minute for three years and ten months. This paper presents an approach to manage a multivariate analysis, in this case merging the values of sensors, and optimizing the observable states to insert into a HMM model, which permits to identify three hidden states that characterize the equipment’s health state: “Proper Function”, “Alert state”, and “Equipment Failure”. The research described in this paper demonstrates how an equipment health diagnosis can be made using the HMM, through the collection of observations from various sensors, without information of machine failures occurrences. The approach developed demonstrated to be robust, even the complexity of the system, having the potential to be generalized to any other type of equipment.