MOLA: Enhancing Industrial Process Monitoring Using Multi-Block Orthogonal Long Short-Term Memory Autoencoder

TL;DR

The paper tackles fault detection in large-scale, nonlinear, and highly integrated industrial processes. It proposes MOLA, which combines an Orthogonal LSTM Autoencoder (OLAE) to learn nonredundant dynamic latent features, a multi-block structure to capture subsystem-specific information, and dual monitoring via Hotelling's $T^2$ on latent codes and a quantile-based multivariate CUSUM, all fused adaptively with Bayesian weights. Key contributions include the OLAE architecture, the quantile-based CUSUM for distribution-shift detection, and the adaptive W-BF fusion to accelerate and robustify detection, demonstrated on the Tennessee Eastman Process with superior fault detection rates and earlier alerts for challenging faults. The approach offers scalable, real-time process monitoring for complex industrial systems and points to future work on data-driven, correlation-aware block partitioning to further enhance robustness.

Abstract

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 $T^2$ statistics and quantile-based cumulative sum (CUSUM) designed for multivariate data streams that are nonparametric, heterogeneous in nature. Compared to having a single model accounting for all process variables, such a multi-block structure improves the overall process monitoring performance significantly, 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, with the goal of improving fault detection speed by assigning weights to blocks based on the sequential order where 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.

Figures (9)

• Figure 1: Illustration of autoencoder structure and feature extraction.
• Figure 2: Illustration of LSTM unit architecture featuring forget, input, and output gates.
• Figure 3: Our proposed OLAE architecture consists of an LSTM encoder, an LSTM decoder, and a fully connected (FC) layer that leverages orthogonality. The FC layer is denoted as $C=\sigma(Wh+b)$, where $h$ and $C$ represent the output of the encoder and fully connected (FC) layer, respectively. Here, $b$ and $\sigma$ are the bias term and nonlinear activation function of the FC layer, respectively.
• Figure 4: The MOLA process monitoring framework features an OLAE model for each block and adaptive W-BF for data fusion.
• Figure 5: Process flow diagram of the TEP.