Predictive Maintenance: Summary of Three Key Research
Studies
1. Predictive Maintenance at Electronic Equipment – A Brief Review
Source: IJTPE, 2015
This paper introduces the core concepts of Predictive Maintenance (PdM) for electronic
systems. It emphasizes that PdM is a proactive strategy relying on continuous or periodic
monitoring of system conditions to detect degradation trends or early-stage faults before they
lead to failures. PdM falls under Condition-Based Maintenance (CBM) strategies and contrasts
with preventive and corrective maintenance by being data-driven and event-based rather than
time-interval-based.
Key Concepts:
PdM involves measuring operational parameters such as temperature, vibration, current,
and voltage to assess equipment health.
Maintenance schedules are generated based on fault trends and historical data rather than
fixed intervals.
PdM can reduce unnecessary maintenance, improve safety, and prevent catastrophic
failures.
Case Studies:
1. Inverter Fault Diagnosis:
An expert system is built using fault-symptom trees for a single-phase inverter. The
system identifies specific inverter faults based on symptoms like voltage absence,
overheating, abnormal control response, or overvoltage. The diagnosis is rule-based and
suitable for microcontroller implementation.
2. DC Motor Monitoring:
A DC motor equipped with sensors and a microcontroller is monitored in real-time.
Temperature, vibration, and voltage/current parameters are measured to detect faults. The
system enables both online and offline analysis and supports remote diagnostics.
2. Predictive Maintenance for Smart Industry
Source: MSc Thesis – İzmir Institute of Technology, 2020
This thesis explores how Machine Learning (ML) can be applied to PdM, particularly using
Adaptive Random Forests (ARF) in a streaming environment. The study uses NASA’s
Turbofan Engine Degradation Simulation Datasets to estimate the Remaining Useful Life
(RUL) of jet engines.
Key Contributions:
Describes three maintenance strategies: Run-to-Failure (R2F), Preventive Maintenance
(PvM), and Predictive Maintenance (PdM).
Highlights the transition from batch learning to stream learning, which processes data
in real-time and adapts to concept drift (changing data distributions over time).
ARF is shown to be effective in PdM because it updates incrementally with each new
data instance.
Methodology:
The dataset simulates multiple jet engines, each running until failure.
Sensors measure variables like temperature, pressure, and vibration over time.
The model predicts RUL based on past data using regression techniques.
Evaluation metrics include RMSE, MAE, and predictive error tracking.
Findings:
Stream learning techniques (like ARF) perform competitively with traditional batch
learning methods.
ARF adapts well to dynamic environments, which is important for real-world industrial
applications.
3. Predictive Maintenance of Induction Motors Using Deep Learning
Source: MSc Thesis – KTH Royal Institute of Technology, 2022
This work focuses on predictive maintenance of 3-phase induction motors, widely used in
manufacturing. The study applies two Deep Learning approaches:
1. Anomaly Detection using Autoencoders
2. Fault Classification using Convolutional Neural Networks (CNNs)
Motivation & Background:
With the rise of Industry 4.0 and IoT, sensors now collect vast amounts of data from
machines.
PdM helps reduce downtime, cost, and increases safety and productivity.
Induction motors are robust but still prone to faults like bearing failure, misalignment, or
electrical imbalance.
Methodology:
Two datasets were used: a public Case Western Reserve University (CWRU) dataset and
a custom Ametller dataset with real sensor data.
Time-series vibration data was transformed into 2D images for CNN input.
Autoencoder models were trained to reconstruct normal signals and detect deviations
(anomalies) indicating faults.
Key Results:
Autoencoders showed high sensitivity in detecting anomalies.
CNN models achieved high accuracy in classifying fault types.
Feature engineering and sensor placement were critical for accurate diagnosis.
Challenges & Limitations:
High-quality labeled data is often scarce.
Implementation in real-time embedded systems (e.g., on ESP32) remains challenging due
to computational complexity.
Overall Insights from All Studies
Theme Summary
Core Idea PdM relies on monitoring operational data to detect faults before failure.
Methods Used Rule-based expert systems, ARF (ML), Autoencoders, CNNs (DL).
Target Equipment Inverters, DC motors, Induction motors, Jet engines.
Common Sensors Vibration, temperature, voltage, current, pressure.
ML/DL Benefits Enables early fault detection, improves scheduling, extends equipment life.
Challenges Requires large, clean datasets and robust real-time processing capabilities.
