Manifold regularized stacked autoencoders-based feature learning for fault detection in industrial processes

The self-attention mechanism has advantages in analyzing the internal characteristics of data and capturing local information. Therefore, it is applied to fault detection. Since complex industrial processes usually contain a mixture of properties, such as nonlinearity, dynamics, and non-Gaussianity, this places higher demands on fault detection methods. To further improve the performance of the self-attention mechanism for fault detection in complex industrial processes, an innovative self-attention mechanism integrating the adaptive double subspace (ISA-ADS) method is proposed for fault detection. First, the query matrix, the key matrix, and the value matrix of the self-attention mechanism are designed so that they can better focus on local information through the adaptive sample weight allocation strategy. At the same time, this design reduces the interference of nonlinear and noisy information and amplifies the impact of important fault information in the sample. Second, considering the problem of Gaussian and non-Gaussian distributions of the data and the different degrees of influence of different subspaces on the final process monitoring results, an adaptive Gaussian and non-Gaussian double subspace fault detection model based on the self-attention output matrix is established, which adaptively assigns different weights to different subspaces by determining their importance. The statistics of different subspaces are fused using the Bayesian inference and a new adaptive monitoring statistic is constructed to obtain a more accurate monitoring of the process state. Finally, the ISA-ADS is validated using the Tennessee Eastman (TE) process and the actual polyethylene production process. The results show that ISA-ADS has a low false alarm rate and a high fault detection rate, verifying the effectiveness of its fault detection performance.

Decentralized adaptively weighted stacked autoencoder-based incipient fault detection for nonlinear industrial processes

Design teacher and supervised dual stacked auto-encoders for quality-relevant fault detection in industrial process

Long–short-term memory encoder–decoder with regularized hidden dynamics for fault detection in industrial processes

In this study, a system for faults detection using a combination of Support Vector Data Description (SVDD) with metaheuristic algorithms is presented. The presented approach is applied to a real industrial process where the set of measured faults is scarce. The original contribution in this work is the industrial context of application and the comparison of swarm intelligence algorithms to optimize the SVDD hyper-parameters. Four recent metaheuristics are compared hereby to solve the corresponding optimization problem in an efficient manner. These optimization techniques are then implemented for fault detection in a multivariate industrial process with non-balanced data. The obtained numerical results seem to be promising when the considered optimization techniques are combined with SVDD. In particular, the Spotted Hyena algorithm outperforms other metaheuristics reaching values of F1 score near 100% in fault detection.

With the continuous evolution of modern industrial technology, industrial production has grown progressively complex, necessitating the use of various sensors to measure multiple process variables. However, intricate temporal dependencies and nonlinear relationships between data presented by multivariate sequences pose significant challenges to process fault detection. In response to these challenges, this paper proposes an attention-based gated recurrent unit autoencoder with skip connection (SAGRU-AE) model for monitoring large-scale, nonlinear, and multivariate industrial process faults. SAGRU-AE combines gated recurrent units, multi-head self-attention, and autoencoder to extract features from multivariable time series data efficiently. Concurrently, feature reuse is achieved through the skip connection structure, which improves the accuracy of data reconstruction. Based on the implementation of process data feature extraction and input reconstruction in SAGRU-AE, two statistics have been developed, namely the H2 statistic and square prediction error (SPE) statistic, for fault detection tasks. Ultimately, the feasibility and effectiveness of the proposed algorithms are validated through experimentation on the TE process.

Recently, principal component analysis (PCA) and independent component analysis (ICA) have attracted increasing attention for fault detection in industrial processes. However, industrial processes usually have multimodal characteristics, in which case, the performance of traditional PCA and ICA methods is poor. A fault detection method based on the local entropy double subspace (LEDS) is proposed that effectively supervises the multimodal process with multiple steady modes. The proposed method uses the local probability density (LPD) to eliminate the multimodal characteristics of the data. Furthermore, local entropy (LE) is applied to mine the statistical information associated with the process data. The process variables are classified into Gaussian and non-Gaussian subspaces by the Kolmogorov–Smirnov (KS) test. PCA and ICA models are built in Gaussian and non-Gaussian subspaces, respectively, for fault detection. A Bayesian decision method is employed to transform the detection results into fault probabilities, and the detection results are combined with the final statistical information. This method effectively eliminates the multicentre structure of the data and improves the fault detection performance of traditional PCA and ICA. The proposed method is applied to the Tennessee Eastman multimodal process. The simulation results show that the fault detection rate of LEDS is better than the rates of traditional methods when the false alarm rate is low for most faults.

Fault detection in continuous industrial processes is crucial to ensure production, quality, safety, environmental and economic aspects. Reliable fault detection systems have become even more necessary due to increasingly stringent regulations. The complexity of processes, on the one hand, and the availability of data, on the other, have spread the adoption of the data-driven approach for industrial fault detection. Deep Reinforcement Learning (RL) has recently been applied for this purpose, mainly in rotating machinery and rolling bearings. In chemical processes, most RL applications are in process control. The main contribution of this work lies in proposing a reinforcement learning (RL)-based framework for fault detection in continuous chemical processes. The challenges are related to the large number of variables, nonlinear relationships and changing operational dynamics, to name a few. The classic Tennessee Eastman Process (TEP) benchmark was used as a case study. The results were compared with Principal Component Analysis (PCA), commonly used for chemical process monitoring, dynamic PCA (DPCA), and neural networks. Compared to PCA, the RL approach showed considerably lower missed detection rate (MDR) and time to detection (TTD) while maintaining reasonable false alarm rate (FAR) for all hard-to-detect faults, contributing to early fault detectability. Similar results were obtained with regard to DPCA. Compared to the supervised approach, RL proved to be more flexible regarding model parameterization, which can be very useful for practical aspects of industrial fault detection. The results show that RL-based approaches can be very useful for fault detection in continuous, non-linear, and time-varying systems, such as chemical processes. For example, with regard to adaptive alarm thresholds and the detection of known and unknown faults. The latter represents a major challenge in industrial practice.

In this work, we introduce MOLA, a multi-block orthogonal long short-term memory autoencoder paradigm, to conduct accurate, reliable fault detection of industrial processes. To achieve this, MOLA effectively extracts dynamic orthogonal features by introducing an orthogonality-based loss function to constrain the latent space output. This helps eliminate the redundancy in the features identified, thereby improving the overall monitoring performance. On top of this, a multi-block monitoring structure is proposed, which categorizes the process variables into multiple blocks by leveraging expert process knowledge about their associations with the overall process. Each block is associated with its specific orthogonal long short-term memory autoencoder model, whose extracted dynamic orthogonal features are monitored by distance-based Hotelling’s T2 statistics and quantile-based cumulative sum (CUSUM) designed for multivariate data streams that are nonparametric and heterogeneous. Compared to having a single model accounting for all process variables, such a multi-block structure significantly improves overall process monitoring performance, especially for large-scale industrial processes. Finally, we propose an adaptive weight-based Bayesian fusion (W-BF) framework to aggregate all block-wise monitoring statistics into a global statistic that we monitor for faults. Fault detection speed and accuracy are improved by assigning and adjusting weights to blocks based on the sequential order in which alarms are raised. We demonstrate the efficiency and effectiveness of our MOLA framework by applying it to the Tennessee Eastman process and comparing the performance with various benchmark methods.

A new fault‐relevant KPCA algorithm is proposed. Then the fault detection approach is proposed based on the fault‐relevant KPCA algorithm. The proposed method further decomposes both the KPCA principal space and residual space into two subspaces. Compared with traditional statistical techniques, the fault subspace is separated based on the fault‐relevant influence. This method can find fault‐relevant principal directions and principal components of systematic subspace and residual subspace for process monitoring. The proposed monitoring approach is applied to Tennessee Eastman process and penicillin fermentation process. The simulation results show the effectiveness of the proposed method.

Data-driven approaches are widely applied for fault detection in industrial process. Recently, a new method for fault detection called principal component pursuit(PCP) is introduced. PCP is not only robust to outliers, but also can accomplish the objectives of model building, fault detection, fault isolation and process reconstruction simultaneously. PCP divides the data matrix into two parts: a fault-free low rank matrix and a sparse matrix with sensor noise and process fault. The statistics presented in this paper fully utilize the information in data matrix. Since the low rank matrix in PCP is similar to principal components matrix in PCA, a T2 statistic is proposed for fault detection in low rank matrix. And this statistic can illustrate that PCP is more sensitive to small variations in variables than PCA. In addition, in sparse matrix, a new monitored statistic performing the online fault detection with PCP-based method is introduced. This statistic uses the mean and the correlation coefficient of variables. Monte Carlo simulation and Tennessee Eastman (TE) benchmark process are provided to illustrate the effectiveness of monitored statistics.

Fault detection via local and nonlocal embedding

Modern industrial plants generally consist of multiple manufacturing units, and the local correlation within each unit can be used to effectively alleviate the effect of spurious correlation and meticulously reflect the operation status of the process system. Therefore, the local correlation, which is called spatial information here, should also be taken into consideration when developing the monitoring model. In this study, a cascaded monitoring network (MoniNet) method is proposed to develop the monitoring model with concurrent analytics of temporal and spatial information. By implementing convolutional operation to each variable, the temporal information that reveals dynamic correlation of process data and spatial information that reflects local characteristics within individual operation unit can be extracted simultaneously. For each convolutional feature, a submodel is developed and then all the submodels are integrated to generate a final monitoring model. Based on the developed model, the operation status of the newly collected sample can be identified by comparing the calculated statistics with their corresponding control limits. Similar to the convolutional neural network (CNN), the MoniNet can also expand its receptive field and capture deeper information by adding more convolutional layers. Besides, the filter selection and submodel development in MoniNet can be replaced to generalize the proposed network to many existing monitoring strategies. The performance of the proposed method is validated using two real industrial processes. The illustration results show that the proposed method can effectively detect process anomalies by concurrent analytics of temporal and spatial information.

Sensor data is extremely important to monitor machines at the shop-floor level and its environmental surrounding conditions for condition-based monitoring, machine diagnosis and process adaptation to new requirements. Based on the described scope, self-diagnostics and self-organizing capabilities are core functionalities of any Industrial Wireless Sensor Network (IWSN). In the present work, a simulated case study was developed with the main intent of validating techniques implemented for sensor data diagnosis of error detection and equipment failure. The scenarios explored try to mimic some common situations of a manufacturing environment when dealing with WSNs, where a piece of sensor equipment suddenly stops working or an unpredictable change in the environment leads to faulty data readings. This paper introduces Castalia and describes how it was used to simulate a direct application of an Optical Metrology System on an industrial Resistance Spot Welding process, which is composed of a camera and several luminosity sensors. More specifically, a sensor data validation module was proposed, implemented and used to extend Castalia functionalities.

Self-learning stationary subspace analysis for fault detection of industrial processes with varying operation conditions