Pith. sign in

REVIEW 3 major objections 5 minor 16 references

FaultExplainer: Leveraging Large Language Models for Interpretable Fault Detection and Diagnosis

T0 review · 3 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

Pith's one-line read The paper claims that an LLM grounded in PCA $T^2$ contributions and a process description can generate plausible and often correct root-cause explanations for chemical-process faults, including some faults outside a historical menu.

desk verdict Useful LLM-PCA integration for TEP fault explanation, but the 'unseen fault' generalization claim rests on an evaluation that can't rule out memorization. read the letter →

arxiv 2412.14492 v1 pith:BDLPIGMF submitted 2024-12-19 cs.AI cs.LGcs.SYeess.SY

classification cs.AIcs.LGcs.SYeess.SY
keywords faultdetectionanddiagnosislargelanguagemodelsTennesseeEastmanProcessprincipalcomponentanalysisT-squaredstatisticrootcausemonitoringLLMhallucination
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper argues that a large language model can serve as an interpretable diagnosis layer over a classical statistical detector, and it demonstrates this in FaultExplainer. FaultExplainer feeds an LLM the six process variables that contribute most to a PCA $T^2$ alarm, along with their deviations from normal operation and a textual description of the Tennessee Eastman Process, and asks the model to name likely root causes and explain how they propagate. With the 15 known root causes listed in the prompt, GPT-4o correctly classifies 7 of the 11 PCA-detectable faults and o1-preview correctly classifies 9; when the list is withheld to mimic unseen faults, both models give the correct or a closely related cause in 8 of 11 cases. These results are offered as evidence that natural-language fault detection and diagnosis is feasible, with the caveat that the explanatory quality is bounded by which variables the PCA feature selection happens to surface.

What carries the argument

The central object is the $T^2$ statistic and its additive variable contributions. PCA is trained on normal-operation TEP data retaining 90% of the variance, and a fault alarm fires when six consecutive $T^2$ values exceed an $\alpha=0.01$ threshold. The six variables with the largest contributions, computed as $\mathrm{CONT}_j = \sum_{i=1}^a \frac{t_i}{\lambda_i} p_{j,i} x_j$, are reported with their deviations from normal operation. That list, together with a detailed TEP process description, forms the prompt that grounds the LLM. Two prompt variants are used: one constraining the model to a menu of 15 known root causes, and one asking for free-form reasoning to mimic unseen faults.

What would settle it

Run the same general-reasoning prompt on a variant of TEP with renamed streams and altered fault mechanisms, or on a different industrial simulator whose faults are not in any public corpus; if correct-or-related accuracy falls well below 8 of 11, the unseen-fault claim is not supported.

Watch

Extended reading notes

Core claim

The central claim is that grounding an LLM in measurable evidence suppresses hallucination well enough to make LLM-based fault diagnosis useful. On the paper's own numbers, the stronger model (o1-preview) gets 9 of 11 PCA-detectable faults right when the root-cause menu is given, and both models get 8 of 11 correct-or-related causes in the free-reasoning setting. The qualitative fault reports show the model tracing each of the six feature deviations back to a single mechanism, including the control system's compensating moves. The failures are also informative: when the PCA-selected features have no direct link to the true mechanism, both models produce confident but wrong causal stories. The paper therefore frames its contribution as a first demonstration rather than a complete solution.

Load-bearing premise

The claim that the general-reasoning prompt tests diagnosis of previously unseen faults rests on the assumption that the models have not memorized Tennessee Eastman Process fault signatures from public literature; if they have, the 8-of-11 score measures recall, not reasoning.

Editorial extensions

If this is right

  • An operator can ask a chat interface why an alarm fired and receive a narrative tying specific sensor deviations to a suspected mechanism, rather than only a red line on a $T^2$ plot.
  • Because the LLM is asked to explain all top-contributing features, the explanation exposes when a candidate cause does not fit the data, turning the qualitative report into a check on the detector's feature choice.
  • The gap between GPT-4o and o1-preview indicates that better reasoning models improve diagnosis even when the underlying PCA detector is unchanged.
  • For faults whose driver variables are absent from the PCA feature set, the system will still generate confident but wrong explanations, so the method's ceiling is set by feature selection, not by the language model.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper, the same grounding recipe should transfer to other multivariate monitoring statistics, such as the squared prediction error or dynamic PCA, which might catch the subtle faults that static PCA misses.
  • A human-in-the-loop study of operators using FaultExplainer would test the implicit claim that readable explanations improve response time or trust; the paper itself only evaluates the text, not the user.
  • A cheap consistency check could reduce hallucinations: require the LLM's proposed cause to name at least one top-contributing variable, and flag reports where the cause and the feature list do not overlap.
  • Comparing FaultExplainer against a non-LLM baseline, such as a rule that maps contribution patterns directly to the 15 known causes, would isolate whether the language model adds diagnostic value or merely reformats the statistics.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper presents FaultExplainer, an interactive tool for fault detection, diagnosis, and explanation in the Tennessee Eastman Process (TEP). The system combines PCA-based T2 fault detection with T2 contribution analysis to select the top six contributing process variables; these variables, their deviations from normal operation, and a written description of the TEP are fed to an LLM (GPT-4o or o1-preview) through one of two prompts: a 'Root Causes-Included' prompt that lists 15 known TEP root causes and a 'General Reasoning' prompt that omits this list in order to mimic previously unseen faults. The authors report that on the 11 PCA-detectable faults, GPT-4o and o1-preview correctly classify 7 and 9 faults, respectively, under the root-cause-included prompt, and that under the general reasoning prompt both models identify the correct or a related cause in 8 of 11 faults. Qualitative examples illustrate plausible explanations for Fault 7 and failures on Faults 10 and 13. The main claim is that the general reasoning results demonstrate 'reasonable performance in diagnosing unseen faults.'

Significance. If the findings were fully supported, the paper would make a useful contribution: it demonstrates a concrete architecture for grounding LLM explanations in classical statistical fault detection, provides an interactive interface and an open-source release, and gives readable examples in which the LLM's explanations track the control-system response in a chemically plausible way. The system-level contribution and the qualitative examples are valuable for the process-systems community. However, the paper's headline generalization claim is not currently supported by the evaluation design, and the quantitative results lack the controls and statistical reporting needed to establish the claimed advantage of the LLM-based reasoning component. With additional experiments and a more cautious framing, the contribution could be of interest to Computers & Chemical Engineering readers working on interpretable FDD.

major comments (3)
  1. [Section 5.3, Table 2] The claim that the General Reasoning Prompt 'reflects real-world scenarios where faults may not have been previously encountered' and that the results 'demonstrate capacity to generalize reasoning beyond fixed fault lists' is not supported by the evaluation. The same 15 known TEP faults are used in both prompt settings, so no genuinely unseen fault is ever tested. Because TEP is one of the most widely used benchmarks in process FDD, and the root causes and variable signatures of these 15 faults are documented in Downs and Vogel (1993) and reproduced in many public papers, GPT-4o and o1-preview may have memorized these fault signatures during pretraining. Removing the root-cause list from the prompt does not remove the model's prior knowledge. The 8/11 score in Table 2 is therefore consistent with recall rather than with causal reasoning from the process description and feature deviations. A no-feature control (same prompt with the PCA deviations omitted) and a model with controlled exposure to TEP would be needed before this score can be interpreted as evidence for diagnosing unseen faults.
  2. [Sections 5.2.1 and 5.3.1, Tables 1 and 2] The quantitative evaluation reports exact counts (7/11, 9/11, 8/11) without confidence intervals, repeated runs, or significance testing. LLM outputs are stochastic, and the scoring rubric is loose: aliases are counted as correct in Table 1, and Table 2 accepts 'related' causes as bolded hits. It is unclear how 'related' was judged, whether the judgment was made by the authors or an independent rater, and whether the counts are stable across prompt temperature settings or repeated API calls. The authors should report the mean and variance over multiple runs, the exact scoring rubric, and inter-rater agreement if the rubric involves subjective judgment.
  3. [Sections 4 and 5] The contribution of the grounding components to the reported performance is not established by any ablation or baseline. The paper does not compare the full system against, for example, the LLM with only the process description, the LLM with only the top-six features, a non-LLM classifier trained on the same PCA features, or a PCA-only rule. Without such a comparison, the claim that the PCA contribution analysis and process description 'ground' the LLM and reduce hallucination remains an anecdotal assertion rather than a demonstrated property of the method.
minor comments (5)
  1. [Section 4.2] The prompt naming is inconsistent: the text says 'The second prompt is named Root Causes-Included Prompt' when the preceding sentence describes the root-cause list as the first prompt. The second prompt should be called the General Reasoning Prompt.
  2. [Section 5.1] The statement that six consecutive T2 exceedances correspond to a false alarm rate of 1e-12 relies on the assumption of independent observations, which is unlikely to hold for autocorrelated process time series. This caveat should be stated or the claim should be removed.
  3. [Sections 4.1 and 5.1] The paper does not specify whether 'top contributing features' ranks contributions by absolute value or by signed value. Since the reported top-six lists include both increases and decreases, the ranking criterion should be stated explicitly.
  4. [Highlights and Section 3] There are several typographical errors: 'process decriptions' in the Highlights, 'T 2 statics' in Section 3 should be 'T2 statistics', and 'FaultExplainer' is inconsistently spaced in Section 3.
  5. [Section 4.2] The paper states that all prompts are shown in the supplementary text, but the supplementary material is not included in the arXiv version. For reproducibility, the prompts should be included in an appendix or the supplementary file should be made available.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: PCA features and LLM outputs are not definitionally tied to the fault labels; the unseen-fault concern is contamination, not circularity.

full rationale

FaultExplainer's claimed contribution is not derived from its inputs by construction. The PCA T2 statistic and contribution scores (Eqs. 1-3) are computed from TEP simulation data under normal operating conditions; the LLM is prompted with the resulting feature deviations plus a process description and is asked to generate root-cause hypotheses. The evaluation compares those hypotheses against independently documented fault labels from Downs and Vogel (1993). No parameter is fitted to the test labels, and no output quantity is defined in terms of the input features in a way that forces the classification. In the Root Causes-Included Prompt, the candidate fault list is provided, but the model's choice among candidates is not mathematically forced by the prompt. In the General Reasoning Prompt, the candidate list is withheld, so the model's outputs are not constructed from the ground-truth labels. The alias scoring rule is an evaluation convention that may inflate reported accuracy, but it does not make the diagnosis equivalent to the input features by definition. The main threat to the 'unseen fault' claim—that GPT-4o and o1-preview may have memorized TEP fault signatures from public literature—is a data-contamination or construct-validity concern about what the LLM's response measures, not a circularity in the paper's derivation; the result would remain empirically unforced by design even if contamination were absent. Self-citations (OptiChat, scChat, LLaVA-Med) appear only as related-work mentions and do not carry the central argument. Therefore, no circular step meeting the quoted-evidence standard is present.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The pipeline is a composition of standard PCA monitoring and LLM prompting. The load-bearing assumptions are that the chosen principal components and top-six contribution features contain enough signal for root-cause discrimination, that the LLM reasons from the provided context rather than from memorized answers, and that the 15 well-known TEP faults can stand in for genuinely unseen faults. The paper itself documents the first and third assumptions as limitations.

free parameters (5)
  • PCA variance captured = 90%
    Number of principal components chosen to retain 90% of the variance in normal operating data.
  • T2 false alarm probability alpha = 0.01
    Threshold for the T2 statistic in Equation 2; the paper sets alpha to 0.01.
  • Consecutive anomaly count = 6
    Fault alarm triggered only when six consecutive T2 values exceed threshold.
  • Top contributing features = 6
    Number of variables with largest contributions fed to the LLM.
  • Top root causes requested = 3
    LLM asked to provide and explain the top three possible causes.
assumptions (5)
  • domain assumption TEP data from Rieth et al. (2017) are representative of the process and contain the 15 known faults.
    Used as ground truth and PCA training data; any errors in the dataset propagate into the evaluation.
  • domain assumption PCA and T2 contribution analysis are sufficient to select variables that can discriminate root causes.
    Section 4.1; the paper's own results show this fails for Faults 10 and 13.
  • domain assumption The LLM's responses can be treated as reasoning from the provided prompt rather than from memorized TEP material.
    Section 5.3; no contamination test is reported.
  • standard math Standard PCA assumptions of linearity and Gaussianity hold for TEP monitoring data.
    Section 4.1 and Section 1 acknowledge this limitation.
  • ad hoc to paper The 15 known faults with root causes from Downs and Vogel (1993) are a valid proxy for unseen fault diagnosis.
    The paper explicitly uses these same faults in both prompt settings, so the claim about 'unseen faults' is not actually tested on new faults.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FaultExplainer: Leveraging Large Language Models for Interpretable Fault Detection and Diagnosis." pith.science (2026). https://pith.science/paper/BDLPIGMF

@misc{pith2026241214492,
  author       = {Pith},
  title        = {Pith review of: FaultExplainer: Leveraging Large Language Models for Interpretable Fault Detection and Diagnosis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BDLPIGMF}},
  note         = {Machine review of arXiv:2412.14492}
}
read the original abstract

Machine learning algorithms are increasingly being applied to fault detection and diagnosis (FDD) in chemical processes. However, existing data-driven FDD platforms often lack interpretability for process operators and struggle to identify root causes of previously unseen faults. This paper presents FaultExplainer, an interactive tool designed to improve fault detection, diagnosis, and explanation in the Tennessee Eastman Process (TEP). FaultExplainer integrates real-time sensor data visualization, Principal Component Analysis (PCA)-based fault detection, and identification of top contributing variables within an interactive user interface powered by large language models (LLMs). We evaluate the LLMs' reasoning capabilities in two scenarios: one where historical root causes are provided, and one where they are not to mimic the challenge of previously unseen faults. Experimental results using GPT-4o and o1-preview models demonstrate the system's strengths in generating plausible and actionable explanations, while also highlighting its limitations, including reliance on PCA-selected features and occasional hallucinations.

Figures

Figures reproduced from arXiv: 2412.14492 by the authors.

Figure 1
Figure 1. Interactive Web Interface for FaultExplainer. From top to bottom are the process monitoring window, fault history tracker, and the interactive chat interface. 8 [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗
Figure 2
Figure 2. Overview of the Methods in FaultExplainer 4. Methods To alleviate the hallucinations of the LLMs, we propose an approach by adapting classical approaches for fault detection and feature importance anal￾ysis to enhance fault explanation in the Tennessee Eastman Process (TEP). The approach combines Principal Component Analysis (PCA) for identifying deviations in process data with T 2 contribution analysis to pinpoint … view at source ↗
Figure 3
Figure 3. An illustrative example where both models are able to identify the correct root [PITH_FULL_IMAGE:figures/full_fig_p016_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: An illustrative example where neither model is able to identify the correct root [PITH_FULL_IMAGE:figures/full_fig_p017_4.png]
Figure 4
Figure 4. Figure 4: This lack of precision resulted in an incomplete and unconvincing [PITH_FULL_IMAGE:figures/full_fig_p018_4.png]
Figure 5
Figure 5. Figure 5: The top six feature changes and the explanation of GPT-4o and o1-preview to [PITH_FULL_IMAGE:figures/full_fig_p021_5.png]
Figure 6
Figure 6. Figure 6: An illustrative example where neither model is able to identify the correct root [PITH_FULL_IMAGE:figures/full_fig_p022_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

16 extracted references · 9 canonical work pages

  1. [4]

    Process Safety and Environmental Protection 165, 463–474

    Xfddc: explainable fault de- tection diagnosis and correction framework for chemical process systems. Process Safety and Environmental Protection 165, 463–474. doi: 10.1016/ j.psep.2022.07.019. Hart, S.,

  2. [7]

    bioRxiv , 2024–10

    scchat: A large language model-powered co-pilot for contextualized single-cell rna sequencing analysis. bioRxiv , 2024–10. Mittelstadt, B., Russell, C., Wachter, S.,

  3. [9]

    Pro- ceedings of the National Academy of Sciences of the United States of Amer- ica 116, 22071–22080

    Defi- nitions, methods, and applications in interpretable machine learning. Pro- ceedings of the National Academy of Sciences of the United States of Amer- ica 116, 22071–22080. doi: 10.1073/pnas.1900654116. OpenAI,

  4. [10]

    OpenAI, 2024a

    Gpt-4 technical report . OpenAI, 2024a. Hello gpt-4o. URL: https://openai.com/index/ hello-gpt-4o/. accessed: 2024-12-17. OpenAI, 2024b. Openai o1. URL: https://openai.com/o1/. accessed: 2024-12-17. Peterson, L.,

  5. [14]

    Nature Machine Intelligence 2023 , 1–11URL: https://www.nature.com/articles/s42256-023-00692-8 , doi:10.1038/s42256-023-00692-8

    Explain- ing machine learning models with interactive natural language con- versations using TalkToModel. Nature Machine Intelligence 2023 , 1–11URL: https://www.nature.com/articles/s42256-023-00692-8 , doi:10.1038/s42256-023-00692-8 . Sun, W., Paiva, A.R.C., Xu, P., Sundaram, A., Braatz, R.D.,

  6. [16]

    AIChE Journal 63, 4329–4342

    Distributed output-feedback fault detection and isolation of cascade process networks. AIChE Journal 63, 4329–4342. doi:10.1002/aic.15791. Zhang, S., Qiu, T.,

  7. [17]

    Chen, H., Constante-Flores, G.E., Li, C.,

    doi: 10.1186/1751-0473-3-17 . Chen, H., Constante-Flores, G.E., Li, C.,

  8. [1987]

    International Journal of Man- Machine Studies 27, 221–234

    Simplifying decision trees. International Journal of Man- Machine Studies 27, 221–234. doi: 10.1016/S0020-7373(87)80053-6 . Ribeiro, M.T., Singh, S., Guestrin, C.,

Show all 16 references
  1. [1997]

    Computers & Chemical Engineering 21, S655–S660

    Signed digraph based multi- ple fault diagnosis. Computers & Chemical Engineering 21, S655–S660. doi:10.1016/S0098-1354(97)87577-1 . Venkatasubramanian, V., Rengaswamy, R., Kavuri, S.N., 2003a. A review of process fault detection and diagnosis: Part ii: Qualitative models and ...

  2. [2008]

    IEEE Transactions on control systems technology 16, 799–808

    Parameter estimation-based fault detection, isolation and recovery for nonlinear satellite models. IEEE Transactions on control systems technology 16, 799–808. Kumari, P., Bhadriraju, B., Wang, Q., Kwon, J.S.I., 2022a. A modified bayesian network to handle cyclic loops in root...

  3. [2016]

    ”Why should i trust you?” Explaining the predictions of any classifier, in: Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Min- ing, Association for Computing Machinery. pp. 1135–1144. doi: 10.1145/ 2939672.2939778. Rieth, C.A., Amsel, B...

  4. [2017]

    URL: https://doi.org/10.7910/DVN/6C3JR1, doi:10.7910/DVN/ 6C3JR1

    Additional Ten- nessee Eastman Process Simulation Data for Anomaly Detection Evalu- ation. URL: https://doi.org/10.7910/DVN/6C3JR1, doi:10.7910/DVN/ 6C3JR1. Sacks, E.,

  5. [2019]

    Explaining explanations in AI, in: F AT* 2019 - Proceedings of the 2019 Conference on Fairness, Accountability, and Transparency, Association for Computing Machinery, Inc. pp. 279–288. doi: 10.1145/3287560.3287574. Murdoch, W.J., Singh, C., Kumbier, K., Abbasi-Asl, R., Yu, B.,

  6. [2022]

    Computers & Chemical Engineering 157, 107619

    Inference of faults through symbolic regression of system data. Computers & Chemical Engineering 157, 107619. doi: 10.1016/j.compchemeng.2021.107619. Harinarayan, R.R.A., Shalinie, S.M.,

  7. [2023]

    arXiv preprint arXiv:2306.00890

    Llava-med: Training a large language-and-vision assistant for biomedicine in one day. arXiv preprint arXiv:2306.00890 . 25 Lu, Y.C., Varghese, A., Nahar, R., Chen, H., Shao, K., Bao, X., Li, C.,

  8. [2024]

    arXiv preprint arXiv:2412.00508

    Graph-to-sfiles: Con- trol structure prediction from process topologies using generative artificial intelligence. arXiv preprint arXiv:2412.00508 . Bhakte, A., Chakane, M., Srinivasan, R.,

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.