Data-driven methods for batch data analysis – A critical overview and mapping on the complexity scale

Dynamic Batch Process Monitoring Based on Time-Slice Latent Variable Correlation Analysis

Batch processes are generally characterized by complex dynamics and remarkable data collinearity, thereby rendering the monitoring of such processes necessary but challenging. This paper proposes a data-driven time-slice latent variable correlation analysis-based model predictive fault detection framework to ensure accurate fault detection in dynamic batch processes. The three-way batch process data are first unfolded into the two-way time slice. For each single time slice, process data are mapped to both major latent variables and residual subspaces to deal with the variable-wise data collinearity and extract dominant data information. A measurement status is then determined with a canonical correlation analysis of the major latent variables and correlated variables, using both the time and batch perspectives. Prediction-based residuals are generated, which provide the basis for identifying the property of faults detected, namely, static or dynamic. Based on experiments using a simulated penicillin production and an industrial inject molding process, the proposed monitoring scheme has been proven feasible and effective.

Data-Driven Approach toward Long-Term Equipment Condition Assessment in Sterile Drug Product Manufacturing

A two-stage data-driven methodology for long-term equipment condition assessment in drug product manufacturing is presented with a case study for a commercially operating aseptic filling line. The methodology leverages process monitoring data. Sensor measurements are partitioned using process information and maintenance schedules that are available on different databases. Data is processed to tackle heterogeneity in sources and formats. The data is cleaned to remove the effects of short-term variabilities and to enhance underlying long-term trends. Two approaches are presented for data analysis: first, anomaly detection using independent component analysis (ICA), where clusters of outliers are identified. The frequency and timing of such outliers yield important insights regarding maintenance schedules and actions. The second approach enables condition monitoring using principal component analysis (PCA). Long-term operational baselines are identified and shifts therein are linked with different process and equipment faults. This approach highlights the impact of equipment deterioration on shifting operational data baselines and shows the potential for the combined application of ICA and PCA for equipment condition monitoring. It can be applied within predictive maintenance applications where the installation of new specialized sensors is difficult, like in the pharmaceutical industry.

Integrating Autoencoder and Heteroscedastic Noise Neural Networks for the Batch Process Soft-Sensor Design

Viscosity represents a key product quality indicator but has been difficult to measure in-process in real-time. This is particularly true if the process involves complex mixing phenomena operated at dynamic conditions. To address this challenge, in this study, we developed an innovative soft sensor by integrating advanced artificial neural networks. The soft sensor first employs a deep learning autoencoder to extract information-rich process features by compressing high-dimensional industrial data and then adopts a heteroscedastic noise neural network to simultaneously predict product viscosity and associated uncertainty. To evaluate its performance, predictions of product viscosity were made for a number of industrial batches operated over different seasons. Furthermore, probabilistic machine learning techniques, including the Gaussian process and the Bayesian neural network, were selected to benchmark against the heteroscedastic noise neural network. Through comparison, it is found that the proposed soft-sensor has both high accuracy and high reliability, indicating its potential for process monitoring and quality control.

First Principles Statistical Process Monitoring of High-Dimensional Industrial Microelectronics Assembly Processes

Modern industrial units collect large amounts of process data based on which advanced process monitoring algorithms continuously assess the status of operations. As an integral part of the development of such algorithms, a reference dataset representative of normal operating conditions is required to evaluate the stability of the process and, after confirming that it is stable, to calibrate a monitoring procedure, i.e., estimate the reference model and set the control limits for the monitoring statistics. The basic assumption is that all relevant “common causes” of variation appear well represented in this reference dataset (using the terminology adopted by the founding father of process monitoring, Walter A. Shewhart). Otherwise, false alarms will inevitably occur during the implementation of the monitoring scheme. However, we argue and demonstrate in this article, that this assumption is often not met in modern industrial systems. Therefore, we introduce a new approach based on the rigorous mechanistic modeling of the dominant modes of common cause variation and the use of stochastic computational simulations to enrich the historical dataset with augmented data representing a comprehensive coverage of the actual operational space. We show how to compute the monitoring statistics and set their control limits, as well as to conduct fault diagnosis when an abnormal event is declared. The proposed method, called AGV (Artificial Generation of common cause Variability) is applied to a Surface Mount Technology (SMT) production line of Bosch Car Multimedia, where more than 17 thousand product variables are simultaneously monitored.

A Review of Kernel Methods for Feature Extraction in Nonlinear Process Monitoring

Kernel methods are a class of learning machines for the fast recognition of nonlinear patterns in any data set. In this paper, the applications of kernel methods for feature extraction in industrial process monitoring are systematically reviewed. First, we describe the reasons for using kernel methods and contextualize them among other machine learning tools. Second, by reviewing a total of 230 papers, this work has identified 12 major issues surrounding the use of kernel methods for nonlinear feature extraction. Each issue was discussed as to why they are important and how they were addressed through the years by many researchers. We also present a breakdown of the commonly used kernel functions, parameter selection routes, and case studies. Lastly, this review provides an outlook into the future of kernel-based process monitoring, which can hopefully instigate more advanced yet practical solutions in the process industries.