Explainability for Fault Detection System in Chemical Processes

Georgios Gravanis; Dimitrios Kyriakou; Spyros Voutetakis; Simira Papadopoulou; Konstantinos Diamantaras

cs.LGFeb 18, 2026

Explainability for Fault Detection System in Chemical Processes

Authors: Georgios Gravanis, Dimitrios Kyriakou, Spyros Voutetakis, Simira Papadopoulou, Konstantinos Diamantaras

Organizations: Department of Information and Electronic Engineering, International Hellenic University, Greece · Chemical Process and Energy Resources Institute, Centre for Research and Technology Hellas, Greece · Department of Industrial Engineering and Management, International Hellenic University, Greece

Abstract

In this work, we apply and compare two state-of-the-art eXplainability Artificial Intelligence (XAI) methods, the Integrated Gradients (IG) and the SHapley Additive exPlanations (SHAP), that explain the fault diagnosis decisions of a highly accurate Long Short-Time Memory (LSTM) classifier. The classifier is trained to detect faults in a benchmark non-linear chemical process, the Tennessee Eastman Process (TEP). It is highlighted how XAI methods can help identify the subsystem of the process where the fault occurred. Using our knowledge of the process, we note that in most cases the same features are indicated as the most important for the decision, while insome cases the SHAP method seems to be more informative and closer to the root cause of the fault. Finally, since the used XAI methods are model-agnostic, the proposed approach is not limited to the specific process and can also be used in similar problems.

Figures & tables

Figure 1: Tennessee Eastman Process Piping and Instrumentation Diagram.

Variable	Description	Units	Variable	Description	Units
xmeas 1	A Feed (Stream 1)	kSm 3 /hr	xmeas 27	Component E	Stream 6
xmeas 2	D Feed (Stream 2)	kg/hr	xmeas 28	Component F	Stream 6
xmeas 3	E Feed (Stream 3)	kg/hr	xmeas 29	Component A	Stream 9
xmeas4	Total Feed (Stream 4)	kSm 3 /hr	xmeas 30	Component B	Stream 9
xmeas 5	Recycle Flow(Stream 8)	kSm 3 /hr	xmeas 31	Component C	Stream 9
xmeas 6	Reactor Feed Rate (Stream 6)	kSm 3 /hr	xmeas 32	Component D	Stream 9

Table 1: TEP variables.

Disturbance	Description	Type
IDV 1	A/C feed ratio, B composition constant (Stream 4)	step
IDV 2	B composition, A/C ratio constant (Stream 4)	step
IDV 3	D feed temperature (Stream 2)	step
IDV 4	reactor cooling water inlet temperature	step
IDV 5	condenser cooling water inlet temperature	step
IDV 6	A feed loss (Stream 1)	step

Table 2: Predefined disturbances in TEP.

Figure 2: Heatmap with the most contributing features. The darker the color, the higher the contribution. A columns for IG , B columns for SHAP

Disturbance	Description	type	Part of the process affected
IDV 3	D feed temperature (Stream 2)	step
IDV 4	Reactor cooling water inlet temperature	step
IDV 9	D feed temperature (Stream 2)	random
IDV 10	C feed temperature (Stream 4)	random	Reactor (direct)
IDV 11	Reactor cooling water inlet temperature	random
IDV 14	Reactor cooling water valve	sticking

Table 3: TEP disturbances grouped by the area affected.

Variable	Description
xmeas 9	Reactor temperature
xmeas 21	Reactors’ cooling water temperature
xmv 10	Reactors’ cooling water valve operation

Table 4: Reactor highly affected variables.

Figure 3: IG and SHAP most important features for IDV 11.

feature	IG	SHAP	feature	IG	SHAP	feature	IG	SHAP	feature	IG	SHAP
xmeas_1	-0.48	0.08	xmeas_15	-0.21	-0.24	xmeas_29	-0.20	-0.29	xmv_1	-0.40	-0.32
xmeas_2	-0.19	-0.32	xmeas_16	-0.08	-0.11	xmeas_30	-0.36	-0.19	xmv_2	-0.22	-0.23
xmeas_3	-0.25	-0.33	xmeas_17	-0.20	-0.37	xmeas_31	-0.19	-0.30	xmv_3	-0.74	0.01
xmeas_4	-0.16	-0.16	xmeas_18	0.01	-0.10	xmeas_32	-0.18	-0.32	xmv_4	-1.24	0.39
xmeas_5	-0.16	-0.27	xmeas_19	-0.26	-0.33	xmeas_33	-0.15	-0.31	xmv_5	-0.20	-0.32
xmeas_6	-0.21	-0.25	xmeas_20	-0.15	-0.08	xmeas_34	-0.28	-0.14	xmv_6	-0.17	-0.15

Table 5: Fault 11 IG and SHAP attribution scores after normalization.

Figure 4: IG attributions for IDV 11. Left: most important features based on IG score, right: less important features based on IG score.

Figure 5: Most and less contributing variable behavior plots for Fault 11.

feature	IG	SHAP	feature	IG	SHAP	feature	IG	SHAP	feature	IG	SHAP
xmeas_1	-0.27	0.98	xmeas_15	-0.37	-0.22	xmeas_29	0.80	-0.03	xmv_1	-0.60	-0.77
xmeas_2	-0.20	-0.39	xmeas_16	0.93	0.33	xmeas_30	-0.31	-0.31	xmv_2	-0.34	-0.32
xmeas_3	-0.28	-0.34	xmeas_17	-0.29	-0.40	xmeas_31	0.57	-0.18	xmv_3	-0.79	0.15
xmeas_4	-0.26	-0.36	xmeas_18	0.27	-3.19	xmeas_32	-0.19	-0.59	xmv_4	-1.40	0.48
xmeas_5	-0.20	0.18	xmeas_19	-0.18	-0.43	xmeas_33	-0.15	-0.37	xmv_5	-0.18	-0.41
xmeas_6	-0.31	0.32	xmeas_20	-0.18	5.04	xmeas_34	-0.40	-0.06	xmv_6	0.38	-0.22

Table 6: Fault 8 IG and SHAP attribution scores after normalization.

Figure 6: IG and SHAP most important features for IDV 8

Figure 7: IDV 8 most important variables according to IG and SHAP

Figure 8: IDV 8 most important variables according to IG and SHAP

Industrial Fault Diagnosis

See all 27 papers

Failure Detection in Chemical Processes Using Symbolic Machine Learning: A Case Study on Ethylene Oxidation
Mar 6, 2026Julien Amblard, Niklas Groll, Matthew Tait +3Fault Detection Interpretable ML
Department of Computing, Imperial College London, London, United Kingdom · Department of Chemical and Biochemical Engineering, Technical University of Denmark, Kongens Lyngby, Denmark · ILASP Limited, London, United Kingdom
BREAD: Baseline-Referenced Explanations for Anomaly Diagnosis
Aug 11, 2026Jiaqi Qiu, Rob Goedhart, Jannis Kurtz +1Feature Attribution Explainable Artificial Intelligence
Department of Business Analytics, Amsterdam Business School, University of Amsterdam, Amsterdam, The Netherlands
TabSHAP
Apr 22, 2026Aryan Chaudhary, Prateek Agarwal, Tejasvi AlladiFeature Attribution SHAP Feature Attribution
Department of Computer Science Birla Institute of Technology and Science India

Mar 6, 2026J/K move · Enter open · S save

Failure Detection in Chemical Processes Using Symbolic Machine Learning: A Case Study on Ethylene Oxidation

Julien Amblard, Niklas Groll, Matthew Tait, Mark Law+2

Department of Computing, Imperial College London, London, United Kingdom · Department of Chemical and Biochemical Engineering, Technical University of Denmark, Kongens Lyngby, Denmark · ILASP Limited, London, United Kingdom

Fault Detection Interpretable ML Manufacturing

Open paper AskPDF arXiv

Over the past decade, Artificial Intelligence has significantly advanced, mostly driven by large-scale neural approaches. However, in the chemical process industry, where safety is critical, these methods are often unsuitable due to their brittleness, and lack of explainability and interpretability. Furthermore, open-source real-world datasets containing historical failures are scarce in this domain. In this paper, we investigate an approach for predicting failures in chemical processes using symbolic machine learning and conduct a feasibility study in the context of an ethylene oxidation process. Our method builds on a state-of-the-art symbolic machine learning system capable of learning predictive models in the form of probabilistic rules from context-dependent noisy examples. This system is a general-purpose symbolic learner, which makes our approach independent of any specific chemical process. To address the lack of real-world failure data, we conduct our feasibility study leveraging data generated from a chemical process simulator. Experimental results show that symbolic machine learning can outperform baseline methods such as random forest and multilayer perceptron, while preserving interpretability through the generation of compact, rule-based predictive models. Finally, we explain how such learned rule-based models could be integrated into agents to assist chemical plant operators in decision-making during potential failures.

Abstract

Industrial Fault Diagnosis

Explore similar work

Failure Detection in Chemical Processes Using Symbolic Machine Learning: A Case Study on Ethylene Oxidation

BREAD: Baseline-Referenced Explanations for Anomaly Diagnosis

TabSHAP