CoChat
Sign Up
Cited by 2,997Machine Fault Diagnosis TechniquesFault Detection and Control SystemsReliability and Maintenance Optimization

The DOI matches its Crossref record; 460 of the 473 references it cites were found in a registry or an index; 3 of the 65 sources its claims cite were checked for support.

Applications of machine learning to machine fault diagnosis: A review and roadmap

Open access
Lei, Yaguo, Yang, Bin, Jiang, Xinwei, Jia, Feng, Li, Naipeng, Nandi, Asoke K.·2020·Mechanical Systems and Signal Processing
DOI 10.1016/j.ymssp.2019.106587
Open this paper in CoChat — free accountOpen PDF

Abstract

This paper presents a review and roadmap to systematically cover the development of intelligent fault diagnosis following the progress of machine learning theories and offer a future perspective.

Overview

AI snapshot

Written from the paper’s full text.

TL;DR

This review traces intelligent fault diagnosis through three eras—traditional machine learning, deep learning, and transfer learning—and proposes a roadmap toward machine health management. It argues that transfer learning is the key future direction for applying diagnosis models in engineering settings where labeled fault data are scarce.

Key findings

  • The authors divide intelligent fault diagnosis into three periods: traditional machine learning before the 2010s, which relied on handcrafted features; deep learning since the 2010s, which enables end-to-end feature learning; and transfer learning as the future direction for cross-task and cross-domain diagnosis.
  • Publication counts from Web of Science show rapid growth since the 2010s, with output from 2017 to 2019 nearly matching the total from before 2017.
  • Among the transfer-learning studies reviewed, most address transfer within the same machine (TIM), while only four publications target transfer across different machines (TDM).
  • The roadmap highlights future challenges including data quality and cleaning, imbalanced health states, streaming data analytics, negative transfer and source-domain selection, and interpretability of deep diagnosis models.

Methods

  • Narrative literature review and roadmap, not an experimental study. It organizes intelligent fault diagnosis by three periods and by method families: expert systems, ANN, SVM, kNN, PGM, and decision trees; stacked AE, DBN, CNN, and ResNet; and transfer-learning categories.
  • Uses Web of Science search results to count publications and motivate the periodization, then reviews representative applications across bearings, gears, motors, engines, and other machines.
  • For transfer learning, it defines domains and tasks, distinguishes TIM from TDM scenarios, and categorizes approaches as feature-based, GAN-based, instance-based, and parameter-based.

Limitations

  • The paper states no limitations of its own review; the following are evident from the text.
  • Evident: no formal systematic-review protocol, search strings, inclusion/exclusion criteria, or quality appraisal are reported, and the publication-counting procedure is described only briefly.
  • Evident: the roadmap and future directions are conceptual and perspective-based rather than empirically tested in the paper.
  • Evident: coverage is selective and organized around the authors’ three-period framework, so it may not exhaustively capture all intelligent fault diagnosis literature.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1016/j.ymssp.2019.106587

Cites

460 of 460 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

3 of 65 cited sources checked · 3 supported · 0 flagged

  • The quote only partly states this claim. 1
  • It may state two separate claims as one. 47
  • Probably not a substantial claim. 52
82 of 82

  • The paper relies on this without support.

    In engineering scenarios, machine users would like an automatic method to shorten the maintenance cycle and improve the diagnosis accuracy.

    “In engineering scenarios, however, the machine users would like an automatic method to shorten the maintenance cycle and improve the diagnosis accuracy.”

  • The paper relies on this without support.

    Traditionally, the process of fault diagnosis is mostly developed by manually inspecting the health states of machines, which increases the labor intensity and reduces the diagnosis accuracy.

    “Traditionally, the process of fault diagnosis is mostly developed by manually inspecting the health states of machines, which increases the labor intensity and reduces the diagnosis accuracy.”

  • The paper relies on this without support.

    Advanced signal processing methods greatly rely on specialized knowledge that maintainers mostly lack in engineering scenarios.

    “However, these methods greatly rely on the specialized knowledge that maintainers mostly lack in engineering scenarios.”

  • The paper relies on this without support.

    The diagnosis results by signal processing methods are too specialized to be understood by the machine users.

    “Furthermore, the diagnosis results by signal processing methods are too specialized to be understood by the machine users.”

  • The paper relies on this without support.

    The volume of collected data is dramatically increasing due to rapid development of internet technologies and IoT.

    “With the rapid development of internet technologies and internet of things (IoT), the volume of collected data is dramatically gathered than ever before.”

  • The paper relies on this without support.

    Increasingly-grown data bring more sufficient information to machine fault diagnosis, making it more possible to provide accurate diagnosis results.

    “The increasingly-grown data bring more sufficient information to machine fault diagnosis so that it is more possible to provide accurate diagnosis results.”

  • The authors state this as an assumption.

    Labeled data are sufficient and contain complete information about the health states of machines.

    “the labeled data are sufficient and contain completed information about the health states of machines [35]”

  • The authors state this as an assumption.

    The source-domain data are labeled, while there are minority of labeled data or even none of labeled data in the target domain.

    “These two categories are both subject to the common assumption, in which the source-domain data are labeled, while there are minority of labeled data or even none of labeled data in the target domain.”

  • The authors state this as an assumption.

    In instance-based approaches, minority of labeled samples are labeled in the target domain, which is insufficient to train reliable diagnosis models.

    “In instance-based approaches, it is assumed that minority of labeled samples are labeled in the target domain, which is insufficient to train reliable diagnosis models.”

  • The authors state this as an assumption.

    In parameter-based approaches, there are minority of labeled samples in the target domain.

    “Similar to the instance-based approaches, parameter-based approaches also assume that there are minority of labeled samples in the target domain [27].”

  • The authors state this as an assumption.

    The successes of transfer learning are subject to the assumption of related health information across multiple diagnosis tasks.

    “The successes of transfer learning are subject to the assumption of related health information across multiple diagnosis tasks.”
11 of 11

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Polynomial-kernel study is assigned the GAN reference number

    Minor

    The discussion attributes a polynomial-kernel MMD study to reference [443], but [443] is a paper on generative adversarial nets. The polynomial-kernel paper appears instead as reference [444].

  • Motors and generators are both allowed and ruled out as a transfer pair

    Minor

    The paper gives transfer between motors and generators as an example of a TDM scenario, but elsewhere says using knowledge from motors to diagnose generators is unrealistic.

2 of 2

Citations

Cites

Flagged first, then by citations

Show all 473 references

13 of the 473 references couldn’t be matched to a known paper.

Cited by

Sorted by citations

See all 2,997

Related

See all

Metadata

Authors
Lei, Y., Yang, B., Jiang, X., Jia, F., Li, N., & Nandi, A. K.
Affiliations
  • Xi'an Jiaotong University (CN)
  • Brunel University of London (GB)
Corresponding author
Lei, Yaguo
ORCID
Venue
Mechanical Systems and Signal Processing
Publisher
Elsevier BV
Published
2020-04
Volume
138
Pages
106587
ISSN
0888-3270 (ISSN-L)
DOI
10.1016/j.ymssp.2019.106587
OpenAlex
W2998506103
Cited by
2,997
References
473
Citations by year
2026: 553 · 2025: 549 · 2024: 497 · 2023: 515 · 2022: 439 · 2021: 327 · 2020: 115
FWCI
172.43
Type
journal article
Language
en
Open access
Licence
Locations
  • Mechanical Systems and Signal Processing (publishedVersion)
  • Brunel University Research Archive (BURA) (Brunel University London) (submittedVersion)
Funders
  • National Natural Science Foundation of China
  • Ministry of Science and Technology of the People's Republic of China
Topics
  • Physical Sciences › Engineering › Control and Systems Engineering › Machine Fault Diagnosis Techniques
  • Physical Sciences › Engineering › Control and Systems Engineering › Fault Detection and Control Systems
  • Physical Sciences › Engineering › Safety, Risk, Reliability and Quality › Reliability and Maintenance Optimization
Keywords
mechanical fault diagnosis, intelligent fault diagnosis, deep learning, transfer learning, condition monitoring, research trends

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-10-11, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-11 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review