The document presents a project on using nonlinear time-series analysis and machine learning algorithms to identify and predict machine faults in rotating machinery. Sensor data is collected from a machine fault simulator and analyzed using time-series analysis, FFT, recurrence plots, and a GRU machine learning model to classify data from good bearings, faulty bearings, and rubbing conditions and forecast future failures. The analysis identified a faulty bearing as the primary cause of deterioration, while the machine learning model validated the need for prompt intervention to address issues and enable proactive maintenance.
OSVC_Meta-Data based Simulation Automation to overcome Verification Challenge...
Machine Fault Prediction Using Time-Series Analysis and ML
1. A PROJECT PRESENTATION
on
NONLINEAR TIME-SERIES ANALYSIS ON MACHINE FAULTS AND
PREDICTION OF FAILURES USING MACHINE LEARNING ALGORITHM
Submitted By
NAME : RAGUL GANDHI R
COURSE : M.E.CAD/CAM ENGINEERING
REG.NO : 210421402011
Under The Guidance of
Dr.M.D.VIJAYAKUMAR M.E,Ph.D.
2. ABSTRACT
The rapid technological advancements have led to the integration of machines
in various industries, enhancing efficiency and productivity.However,
machine faults and failures can cause significant downtime and financial
losses.
Nonlinear time series analysis and machine learning algorithms offer
promising ways to identify and predict machine faults proactively,
minimizing downtime and enabling proactive maintenance.
A machine fault simulator acquires vibrational signals for different scenarios,
and nonlinear time series analysis, such as Recurrence plots and FFT
computation, is applied to characterize the signal.
Machine learning algorithms utilize historical sensor data to build
predictive models. By identifying fault-associated patterns, these algorithms
accurately predict potential future failures.
3. INTRODUCTION
The Present Experiment proposes a machinery fault kit designed to detect faults in
bearings, distinguish between good and faulty bearings, and rubbing attachments.
Data is collected using sensors and a Data Acquisition System (DAQ), and then stored
for analysis. The analysis includes time series analysis for the three conditions (faulty
bearing, good bearing, and rubbing attachment), as well as Fast Fourier Transform (FFT)
to identify system natural frequencies and extreme events.
To understand amplitude patterns during operation, recurrence plots are generated to
visualize repetitive patterns.
Additionally, a machine learning algorithm, specifically the Reservoir Computing Gated
Recurrent Unit (GRU) method, is utilized for forecasting future failures and predicting.
potential issues.
4. VIBRATION ON ROTATING
MACHINES
Vibration in rotating systems is oscillatory motion caused by imbalances,
misalignments, or external forces.
It can be free (undisturbed) or forced (externally influenced).
Excessive vibration can be detrimental to the rotating system. It can
lead to increased wear and tear on components, reduced overall
performance, and in extreme cases, mechanical failure.
To minimize the adverse effects of vibration, engineers employ various
techniques: Engineers use balancing, alignment, and damping to
minimize vibration effects.
5. METHODOLOGY
Machine
Specification
Mounting Sensor
on Machine
Data Acquisition
System(DAQ)
Computer
Collected Data
Analysis
Time Series Analysis
FFT
Recurrence Plots
Finding Extreme
Event
Forecasting (ML)
Identification of
Defect
Classification of
Defect
Decision Making
6. Machine fault simulator speed controller
Torque controller
Time domain signals and frequency
domain signals of vibration
Total data sets(collected
via DAQ)
TRAINING
DATA
TESTING
DATA
BUILDING
TRAINED
DECISION
MAKING
80%
20%
8. DATAACQUISITION
DAQ stands for Data Acquisition, which refers to the process of capturing,
measuring, and converting real-world physical signals into digital data for further
processing, analysis, and storage.
The Data Acquisition (DAQ) system with two channels is essential for collecting
vibration data using accelerometers, which measure the acceleration of a
vibrating body.
Two accelerometers are mounted at strategic points on the vibrating body (0 and
1) to capture initial vibrations near the bearings.
Once the analog signals are converted into digital data, they are either stored in a
data buffer or transmitted directly to a computer for further processing and
analysis.
1. MAX SAMPLING FREQUENCY (KHz) 48
2. ADC Resolution(bit) 24
3. Input Voltage Range(V) 10
4. FREQUENCY RESPONSE (Hz) 1-20000
9. ACCERELOMETER
The Accelerometer sensor used in this setup offers a wide frequency response
range, spanning from 30 to an impressive 900,000 CPM (Cycles Per Minute).
This wide frequency range allows the sensor to capture vibrations across a broad
spectrum of frequencies, making it suitable for various applications where different
vibration frequencies may be present.
Whether the vibrations are relatively slow or high-speed, this accelerometer can
effectively measure and record the acceleration data.
Furthermore, the sensor is designed to withstand extreme temperature conditions,
with a wide operating temperature range from -50°C to 121°C.
This temperature range is particularly advantageous for applications in harsh
environments or industrial settings where temperature variations can be significant.
10. BEARING
In a rotating system, a bearing is a mechanical component that supports the rotation of a shaft or
axle. Bearings are essential for reducing friction and facilitating smooth movement between two
moving parts.
A bearing fault in a rotating system refers to any abnormal condition or damage that occurs in the
bearing, leading to a decrease in its performance or potential failure. Bearings can experience
various types of faults due to
1. WEAR
2. FATIGUE
3. LUBRICATION ISSUES
4. MISALIGNMENT
5. CORROSION
11. INNER AND OUTER RACE
BEARING FAULT & RUBBING
An inner race fault in a bearing refers to a specific type of fault that
occurs on the inner ring or raceway of the bearing. The inner race is the
ring that is mounted on the rotating shaft of the machinery. Faults on the
inner race can occur due to various factors, such as excessive loading,
misalignment, improper lubrication, or contamination.
An outer race fault in a bearing refers to a type of fault that occurs on the
outer ring or raceway of the bearing. The outer race is usually mounted
in a stationary housing. Outer race faults can be caused by factors
similar to inner race faults
In a rotating system, rubbing refers to the contact or interaction between
two components that are not meant to touch or come into contact with
each other during normal operation. .
12. Time series analysis is a statistical technique used to analyze and interpret data that
is collected over time. In the context of a rotating system and bearing fault
detection,time series analysis involves studying the behavior and patterns of
vibration data. It helps in identifying trends, periodicities, and abnormalities that
may indicate the presence of bearing faults or other issues in the rotating system.
ANALYSIS
1. TIME SERIES ANALYSIS
2. FFT(FAST FOURIER TRANSFORM)
3. EXTREME EVENT
4. RECURRENCE
5. FORECASTING(MACHINE LEARNING)
13. FFT is a mathematical algorithm used to transform time-domain data into the
frequency domain. In the context of rotating systems, FFT is applied to the time
series data, such as vibration signals, to analyze the system's natural frequencies and
frequencies associated with different fault modes
Extreme Events in rotating system refer to abnormal or anomalous occurrences that
deviate significantly from normal operating conditions. These events can be caused
by bearing faults, imbalances, misalignments, or other factors. Extreme event
analysis involves identifying and characterizing these rare events in the time series
data
Recurrence plots are graphical representations used to visualize and analyze
repetitive patterns in time series data. In the context of rotating systems, recurrence
plots can help in identifying periodicities or cyclic behaviors in the vibration or
sensor data.
Machine learning techniques, such as the Reservoir Computing Gated Recurrent
Unit (GRU) method mentioned earlier, can be employed for Forecasting future
failures in rotating systems.
20. RESULTS
Faulty bearing identified as the primary cause of machine deterioration.
Rubbing condition found to have minimal impact but not neglected due to
potential consequences.
Observed deviations approaching critical levels, urging immediate action.
Advanced Machine Learning techniques validated the need for prompt
intervention.
Good bearing demonstrated reliability and stability, not contributing to
issues.
Timely addressing of faulty bearing and rubbing concerns is crucial.
Monitoring and maintaining good bearing's performance emphasized.
Proactive maintenance strategies necessary to prevent costly downtime
and failures.
Thorough analysis and ML predictions provide a foundation for informed
decision-making.
21. RECOMANTATION
Ensure accurate and well-calibrated sensors for reliable data collection.
Implement a robust data storage and management system for easy data
access.
Automate data analysis to streamline fault detection and save time.
Train the ML algorithm on diverse datasets and update it regularly.
Implement real-time monitoring for timely anomaly detection.
Provide clear visualizations and reports for easy interpretation.
Schedule proactive maintenance based on ML forecasting.
Gather feedback for continuous improvement of the fault detection system.
Thoroughly validate the fault kit's capabilities across various conditions.
Document all procedures and conduct training for personnel.
25. The Project was Displayed at ACMEE 2023, the 15th
International Machine Tools Exhibition
26. REFERENCE
1. R. H., & Stoffer, D. S. (2017). Time Series Analysis and Its Applications: With R Examples
(4th ed.). Springer.
2. Oppenheim, A. V., & Schafer, R. W. (2010). Discrete-Time Signal Processing (3rd ed.).
Pearson.
3. Embrechts, P., Klüppelberg, C., & Mikosch, T. (2013). Modelling Extremal Events: For
Insurance and Finance (2nd ed.). Springer.
4. Marwan, N., Romano, M. C., Thiel, M., & Kurths, J. (2007). Recurrence Plots for the
Analysis of Complex Systems. Physics Reports, 438(5-6), 237-329.
5. Hyndman, R. J., & Athanasopoulos, G. (2018). Forecasting: Principles and Practice (2nd
ed.). OTexts.
6. A Novel Approach to Predictive Maintenance in Rotating Machinery using Machine
Learning Techniques