21MET807

Machinery Fault Diagnostics and

Signal Processing

Dr. Naresh Kumar Raghuwanshi

Assistant Professor
Department of Mechanical Engineering
Malaviya National Institute of Technology Jaipur
 Machinery Condition Monitoring
It deals with the maintenance aspects of machines based on the present and
past condition of the machine.

It is the process of monitoring a parameter of condition in machinery

(vibration, temperature etc.), in order to identify a significant change which is
indicative of a developing fault.

The use of condition monitoring allows maintenance to be scheduled, or

other actions to be taken to prevent any catastrophic failure of the machine.

The majority of machines have rotating components.

 Future Needs

New techniques for fault diagnosis in machines are always emerging with
the advancement of technology.

There has been improvement in all aspects of condition monitoring, such as

transducers and instrumentation, digital signal processing, and decision
making.

The use of wireless systems for data transfer from the measurement location
to the analysis station is preferred.

Now a days IOT is playing an important role.

Machine Condition Chart
 Types of Machine Maintenance in Plants

• Preventive Maintenance:
In preventive maintenance, the maintenance on a particular machine is done in a regular periodic manner at a
fixed frequency.
Also known as periodic maintenance
Time-Based Maintenance (TBM):
Periodic-based maintenance, uses a calendar schedule that is set in advance and performs maintenance
regardless of the health of the machine, which makes this approach expensive for some large, complex machines.
In addition, TBM may not prevent machine failures.

• Predictive Maintenance:
Predictive maintenance is a need-based maintenance that depends on the condition of the machine.
The decision to conduct maintenance or not depends upon the machine’s past and present condition.
Also known as Condition-Based Maintenance (CBM).

• Reactive Maintenance:
Reactive maintenance means that we react to a need for maintenance of the machine, because the machine has
failed completely.
Another name for reactive maintenance is breakdown maintenance or Corrective Maintenance .
Normally, based on the designer’s input, the original equipment manufacturer
provides a schedule for the periodic maintenance to be done on the machine.

For example, new vehicle requires preventive maintenance as per the schedule
mentioned in the service manual of the vehicle supplied by the manufacturer at a
fixed time interval or distance covered by the vehicle.
The service could be for an engine oil and filter change or some other part
replacement if any problem.

Another example where preventive maintenance is used is for strategic reasons. In

the armed forces, we would always like the weapons to fire whenever required to
do so, army vehicles and naval vessels to move immediately on an order from the
field commander, and so on. In such scenarios, cost is not a major criterion but
functionality of the system is.
Predictive Maintenance (condition-based maintenance)
Predictive maintenance is a need-based maintenance that depends on the condition of the machine.
Also known as condition-based maintenance.
The decision to conduct maintenance or not depends upon the machine’s past and present condition.

For this additional instrumentation is required on the machine to measure its “health” parameters.
Instrumentation includes transducers, signal conditioners, data acquisition units, computer-based signal
analysis systems, and software-driven diagnostic routines.

Vibration-based condition monitoring includes investigating the reason for the frequent and repetitive
failures in any machines by doing either additional vibration tests or further analysis of the measured
vibration data in addition to the regular machine condition monitoring.
Advantages of the predictive maintenance:
• Lower maintenance costs
• Fewer machine failures
• Less repair downtime
• Reduced small parts inventory
• Longer machine life
• Increased production
• Improved operator safety

Major disadvantages of predictive maintenance:

Requires extra investment in the initial stage for the additional instrumentation
and it needs a robust software for fault diagnosis and prognosis.
Bath Tub Curve
Infant mortality zone (Machine Installation)
High failure rates occurs in the early stages of the machine. Reasons for such high failure rates, faulty installation at the site,
ignorance and unfamiliarity of the machine operator, improper electrical power supply, non-availability of a user or training manual,
improper specifications, and choice of the machine.
Useful period (Random Failure)
Once the above reasons are sorted out, the machine’s failure rate reduces significantly; this state of the machine continues for a
considerable time, which is known as the useful life of the machine. It is related to Random Failure.
Wear-out zone (Frequent Failure )
The failure rate again increases, which can be due to excessive wear and tear and fatigue failure of the machine component. Though
by maintenance the failure rates can be controlled and reduced, a time comes when the cost of maintenance or upkeep is so high
that it is better to completely replace the machine with a new one.

Availability of a Machine =
Useful period (also known as uptime)

To increase the availability of a machine by decrease the

machine’s downtime.

Total lifespan = Uptime + Downtime

Some definitions
Fault: another word for a problem.
Fault Detection:
Fault detection is recognizing that a problem has occurred, even if you don't yet know the root cause.

Faults may be detected by a variety of quantitative or qualitative means.

It also includes such as alarms based on high, low, or deviation limits for process variables or rates of

Fault Diagnosis:
Fault diagnosis is pinpointing one or more root causes of problems, to the point where corrective action can
be taken.
This is also referred to as “fault isolation”, especially when emphasizing the distinction from fault
detection.

Fault Prognosis:
Once the faults have been detected and diagnosed, the next question is how long the machine will last in
the present condition or what is the remaining useful life (RUL). Many deterministic and stochastic
approaches are available for predicting the RUL of a machine.
Condition Monitoring Techniques (Fault Diagnosis Techniques)
• Oil Analysis
• Wear Debris Analysis
• Vibration Analysis (Vibration Monitoring)
• Noise Monitoring
• Motor Circuit Analysis
• Thermography/Temperature Measurements/Infrared Thermography
• Ultrasonic Monitoring/Acoustic Analysis/Airborne Ultrasonics
• Radiography/Radiation Analysis/Neutron Radiography
• Laser Interferometry
• Electrical Monitoring
• Electromagnetic Measurement/ Eddy current
 Oil Analysis/Tribology
Oil Analysis applies to machine oils, lubricants and fluids.
It can detect wear, overheating, and contamination.
High levels of iron, for example, often indicate dirt and grit.
This reduces gearboxes failures by 50%.
Avoiding contamination decreases bearing failure by 75%.
There are several oil analysis methods available:
Ferrography,
Water presence tests,
Viscosity tests,
ICP/ spectroscopy,
dielectric strength test,
Iron content (particle quantification index),
Infrared spectroscopy,
Ultraviolet spectroscopy,
Potentiometric titration (total acid number/ total base number),
Sediment tests etc.
 Wear Debris Analysis

Wear debris analysis is one of the key techniques to evaluate

wear particle counts and measure their sizes for the tested
fuels.

This study is conducted using the wear debris analysis by

SEM/EDX etc. on the filtered metal debris samples. The
samples are collected by filtering the oil from the four-ball
tribotester.

Once confidence can be established in the concentration and

quality of the wear metals in the oil, appropriate alarms and
limits can be set.

The most important tool is the microscope, which is used by

an analyst trained in tribology.
Vibration Analysis
It is one of the most well-known predictive maintenance methods.
Different parts vibrate in different ways, and worn or out-of-balance
parts have unique vibration signatures that can be tracked and used
to predict parts failures.

It can detect misalignments, imbalances, and wear about

3 months before they cause a breakdown.

Techniques include:
• Shock pulse analysis
• Fast Fourier transforms
• Broadband vibration analysis
• Ultrasonic analysis
• Power spectral density (PSD)
• Time waveform analysis
• Spectrogram/spectrum analysis
 Noise Analysis
Unwanted (unpleasand) sound is considered as Noise.
Noise signal processing has been attracted considerable attention for fault prevention
and detection systems because the noise signals are not dependent on the power range
and size of rotating machines.
The signal processing is required as in case of vibration analysis.
 Ultrasonic Monitoring/Acoustic Analysis/Airborne Ultrasonics
Ultrasonic monitoring uses high-frequency sound waves to detect leaks, part defects parts seating, and
cavitations etc.

Airborne and structure-borne ultrasound are gaining traction as a good option to detect ‘stress waves’ on
rotating machinery.

It provides an early warning for machine parts

deterioration that might otherwise be masked by
ambient plant noises and temperatures.
In many cases, tiny changes in friction forces can be detected with ultrasonic monitoring. These small
changes may be missed with IR or vibration analysis. Because of this, UM can be an excellent
companion testing technique along with IR and vibration analysis.
Almost all areas of manufacturing processes can benefit from ultrasonic monitoring UM.
 Motor Circuit Analysis (MCA)
MCA test assesses the condition of electric motors.
Electrical imbalances and degradation of insulation are the chief causes of motor failure.

Some tests are generally grouped into voltage-based or current-based tests.

Inspection points include:

• Power circuit/current signature
• Rotor
• Stator
• Insulation
• Power quality
• Air gap
Thermography/Temperature Measurements/Infrared Thermography
Thermography is usually referred to as infrared thermography or thermal imaging.

Thermography is the process of using an infrared camera to look for abnormally hot or cold areas on a component
operating under normal conditions.
Thermography involves the use of thermal camera, which are sophisticated devices that
measure the natural emission of infrared radiation from a heated object to produce a
thermal picture or video.

It is a non-destructive technique (NDT).

Thermography is useful for the detection of corrosion/erosion damage in plants operating at elevated
temperatures.
In addition, it can be used to check for fouling or internal plugging of piping systems and to check the quality of
refractory linings.
It can also be used for leak detection, composition changes and disbonds in laminates.
 Radiography/Radiation Analysis/Neutron Radiography
Like thermography, radiography (including radiation analysis and neutron radiography) is a very
thorough method of non-destructive testing.
This method uses radiation imaging to identify internal defects in equipment and parts.

Applications include inspection of castings and sintered parts, flaws in welding etc.

The technique is based on measuring the differential absorption of radiation penetrating the
part or material.
Internal corrosion and flaws absorb differing amounts of radiation, which can be measured
and analyzed.
Techniques include:
• Neutron backscatter
• Computed radiography
• Computed tomography (CT)
• Direct radiography
• Positive material identification
(PMI)
• Neutron radiography
Laser interferometry

It is based on the interference of light waves generated by a laser.

The interference pattern is then captured and measured by a device called an
interferometer.
The various interference patterns can be analyzed to show differences in material
characteristics such as the presence of corrosion, surface defects.
Electrical Monitoring
This approach applies the principles of deviations in electrical parameters to identify
faults and defects. Characteristics such as resistance, induction, capacitance, pulse
response, frequency response and others are used to detect potential maintenance
issues.

Power signature analysis (to test current and voltage), battery impedance testing, surge
and hipot testing (also to detect insulation flaws) and, to some extent, motor circuit
analysis too.
Electromagnetic Measurement / Eddy-current Testing

Electromagnetic measurement shouldn’t be confused with electrical monitoring.

Electromagnetic monitoring measures magnetic field distortions to spot cracks, dents,
corrosion, weaknesses, and other defects (e.g. thinning).

Perhaps, the most prominent method of electromagnetic condition monitoring is eddy-

current testing (ECT), which is used in the petrochemical industry to detect tight cracks.

Techniques include:
• ECT
• Magnetic particle inspection
• Magnetic flux leakage
• Metal magnetic memory method
• Pulsed eddy currents
• Remote and near field eddy current
• Saturated low-frequency eddy currents
• Other eddy current testing

http://www.ndttechnologies.com/products/eddy_current_testing.html
Artificial Intelligence (AI) is the branch of computer science that focuses on creating systems capable of performing
tasks that typically require human intelligence.

• These tasks include learning from experience, recognizing patterns, understanding and generating natural
language, solving problems, and making decisions.
• AI systems achieve these capabilities through algorithms and models that process data, allowing them to adapt
and improve over time.

Machine Learning (ML) is a branch of AI that focuses on building systems that can learn from and make decisions
based on data.

Instead of being explicitly programmed to perform a specific task, machine learning models are trained using large
amounts of data and algorithms that enable them to identify patterns, make predictions, and improve over time with
more data.
Deep Learning (DL) is a specialized subset of machine learning that uses neural networks with multiple layers to
model and understand complex patterns in data.
• It is inspired by the structure and function of the human brain, where networks of neurons process and transmit
information.
• Deep learning models are particularly powerful for tasks where data is abundant and complex.
Data Science is an interdisciplinary field that combines techniques from statistics, computer science, and domain
expertise to extract meaningful insights and knowledge from data.
• It involves collecting, cleaning, analyzing, and interpreting large volumes of data to solve complex problems, make
data-driven decisions, and predict future trends.
 Types of Learning/Machine Learning
1. Supervised Learning: In supervised learning, the model is trained on a labeled dataset, meaning that each input comes
with an associated output (label). The model learns to map inputs to outputs, and the goal is to make accurate
predictions when given new, unseen data.

2. Unsupervised Learning: Unsupervised learning involves training models on data without labeled outputs. The goal is
to identify hidden patterns or intrinsic structures in the input data.

3. Semi-Supervised Learning: Semi-supervised learning is a machine learning approach that combines a small amount of
labeled data with a large amount of unlabeled data during training. Traditionally, supervised classifiers need only
labelled data to train.

4. Reinforcement Learning: In reinforcement learning, an agent learns to make decisions by interacting with an
environment. An agent is any system that can make decisions and act in response to its environment independent of
direct instruction by a human user. Robots and self-driving cars are examples of autonomous agents.

5. Transfer Learning: Transfer learning involves taking a pre-trained model on one task and applying it to a different but
related task. This approach is especially useful when there is limited data for the new task. In this learning technique,
labelled data from a different domain is used to enhance the learning process.

6. Online Learning (Incremental Learning): Online learning is a method where the model is updated continuously as new
data becomes available, rather than being trained on a fixed dataset all at once.
Dimensionality Reduction

Feature Extraction:
• Feature extraction is a dimensionality reduction procedure that transforms high-dimensional input
into a reduced set of features.
• The feature transformation techniques aim to reduce the high-dimensional input data to a low-
dimensional set that are linear or nonlinear combinations of the original signals.
• Linear subspace learning (LSL)
• Nonlinear subspace learning (NLSL)

Feature Selection:
• Feature selection, also called subset selection, techniques aim to select a subset of features that can
sufficiently represent the characteristic of the original features.
• The main task of feature selection is to select a subset of features that are the most relevant to the
classification problem.
• In view of that, this will reduce the computational cost and may remove irrelevant and redundant
features and consequently improve learning performance.
 Machine Learning based Fault Diagnosis
Learning from Vibration Data
Visual inspection is not dependable in the rotating machine condition monitoring field for many reasons that
can be summarised as follows:
(1) Not all waveform signals from rotating machines provide clear visual differences.
(2) In practice, we deal with a large collection of vibration signals that often contain background noise.
(3) Sometimes we deal with low-amplitude signals measured in a noisy background.

The demand for early detection of faults makes manual inspection of all the collected vibration signals
impractical. Hence, a learning algorithm is required to automatically detect meaningful patterns in the
vibration data.

31
 Machine Learning Algorithms
1. Naive Bayes (NB) machine learning classifier
This classifier is based on Bayes theorem of probability.
Fast and efficient: NB is computationally efficient as it only requires a small amount of training data to
estimate the parameters necessary for classification.
If A and B are two events, then the formula for the Bayes theorem is given by:

Where P(A|B) is the probability of condition when event A is occurring while event B has already occurred.
 Machine Learning Algorithms
2. K-Nearest Neighbors (KNN) machine learning classifier
KNN is a lazy learner machine algorithm.
All training dataset points are used in the phase of testing for fault classification.
It does small work in training of datasets and large amount of work in the testing stage to make a classification
accurately.
It shows better results with a small number of classification features.
When dimension of dataset is increases by increasing classification features generates the issues of over-fitting for
the recorded datasets.

35
 Machine Learning Algorithms (Contd…)
3. Decision Tree (DT) machine learning classifier
As the name indicates, it generates a database model in the form of a tree like
structure.
This algorithm can be used for multi-dimensional analysis with multiple classes
of fault detection.
Feature vector is defined by the each node in the decision tree classifier.
There is over fitting issue is related to this classifier.

Decision Tree Classification

36
 Machine Learning Algorithms (Contd…)
4. Random Forest (RF) machine learning classifier
Random Forest has the maximum flexibility and simple to use as a machine learning algorithm.
It is a combining classifier that combines many decision tree classifiers to overcome the over fitting issues
of the datasets.
This classifier works efficiently on more number of dataset points for fault classification.

37
 Machine Learning Algorithms (Contd…)
5. Support Vector Machine (SVM)
⁕ It is completely based on the concept of surface creation which is called as hyper plane.
⁕ It can create N-dimensional hyperplane that assigns the new data class into one of the two output classes
(healthy and faulty class).
⁕ The key task of an SVM machine algorithm is to prediction a class from which a new upcoming data point fit
in.
⁕ Effective in High-Dimensional Spaces
⁕ Works Well with Non-linear Data
⁕ Handles Large Feature Spaces

38
 Artificial Neural Network (ANN)
An artificial neural network (ANN) is a mathematical model that mimics biological neurons. A neuron is a
special biological cell that process information.

Linear function

Activation function

Biological Neurons Model of an artificial neuron

Input layer Hidden layers Output layer

Multilayer model for ANN