Abstract:Effective fault detection and diagnosis (FDD) is crucial for proactively identifying irregular states that could jeopardize operator well-being and process integrity. In the era of Industry 4.0, data-driven FDD techniques have received particular attention, driven by the proliferation of stored manufacturing sensor data. While these methods have proven adept at categorizing established process fault scenarios, there remains an imperative to identify and explain anomalies stemming from uncharted faults or the interplay of consecutive anomalies. To address this we present a knowledge-enhanced FDD approach that integrates well-defined chemical engineering knowledge with cutting-edge deep learning techniques. We apply our methodology, named Knowledge-Enhanced Spatiotemporal Analysis (KESA), to identify abnormal process conditions that may be a precursor to failure. Furthermore, we utilize the knowledge of the fundamental relationships governing the process to explain why this fault case has occurred. This type of in-depth fault analysis is only possible through leveraging domain expertise and marks a step forward in FDD technology in comparison to current literature. Using the benchmark Tennessee Eastman process dataset, we establish superiority in the accuracy and efficiency of our KESA model against state-of-the-art FDD algorithms. This work highlights the importance of a knowledge-enhanced approach to deep learning in complex environments, emphasizing the critical role of timely and interpretable fault detection. By providing explanations for model results, our KESA framework not only aids in effective decision-making but also has the potential to significantly reduce the time between fault detection and the implementation of proactive mitigation actions. This capability is paramount for improving overall safety, minimizing downtime, and ultimately contributing to substantial cost savings in industrial processes.

ProcK: Machine Learning for Knowledge-Intensive Processes

Multiphase Flow Modeling Using Process Knowledge Integrating Temporal Graph Convolution Network

Nonlinear Process Monitoring Using Improved Kernel Principal Component Analysis

Learning a Data-Dependent Kernel Function for KPCA-based Nonlinear Process Monitoring

Deep Neural Network-Embedded Stochastic Nonlinear State-Space Models and Their Applications to Process Monitoring

A sequential cross-product knowledge accumulation, extraction and transfer framework for machine learning-based production process modelling

Nonlinear Process Monitoring Based on Maximum Variance Unfolding Projections.

Enhanced Batch Process Monitoring Using Just-in-time-learning Based Kernel Partial Least Squares

Knowledge Automation Through Graph Mining, Convolution and Explanation Framework: A Soft Sensor Practice

Process Pattern Construction and Multi-Mode Monitoring

An automatic method for constructing machining process knowledge base from knowledge graph

Predictive Process Model Monitoring using Recurrent Neural Networks

Deep graph kernel point processes

Machine Learning in Business Process Monitoring: A Comparison of Deep Learning and Classical Approaches Used for Outcome Prediction

Predictive process model monitoring using long short-term memory networks

HOEG: A New Approach for Object-Centric Predictive Process Monitoring

Convolutional Neural Networks in Process Mining and Data Analytics for Prediction Accuracy

Knowledge-Enhanced Spatiotemporal Analysis for Anomaly Detection in Process Manufacturing

PDPK: A Framework to Synthesise Process Data and Corresponding Procedural Knowledge for Manufacturing

Can deep neural networks learn process model structure? An assessment framework and analysis

A Process Knowledge-Based Hybrid Method for Univariate Time Series Prediction with Uncertain Inputs in Process Industry