Final Report 
ARTIFICIAL 
INTELLIGENCE 
Artificial intelligence for fault diagnosis of rotating 
machinery: A review 

NAGA SAI RAM AELLA 18BCE1177

VISHNU VARDHAN REDDY M 18BCE1150

SAI RAM TEJA N 18BCE1173

KARTHIK REDDY D 18BCE1203 

Outline 
1. Introduction  

2. Theoretical background of AI approaches 

2.1. k-Nearest neighbour 

2.2. Naive Bayes classifier 

2.3. Support vector machine  

2.4. Artificial neural network 

2.5. Deep learning 

3. Discussion, limitation and future trends. 

4. Conclusion 

Acknowledgment  

References 

1 
 

Introduction 
Fault diagnosis of rotating machinery is a technique of fault detection, isolation and identification, 
which can be applied on the information about the operation condition of the equipment. 

There are three basic tasks of fault diagnosis: 

(1) determining whether the equipment is normal or not; 

(2) finding the failure and its reason;  

(3) predicting the trend of fault development; 

As a powerful pattern recognition tool, artificial intelligence (AI) has attracted great attention from 
many researchers and shows promise in rotating machinery fault recognition applications. 

A common fault diagnosis system often consists of two key steps:  

data processing (feature extraction) 

fault recognition  

Most common intelligent fault diagnosis systems are built based on the preprocessing by feature 
extraction algorithms to transform the input patterns so that they can be represented by 
low-dimensional feature vectors for easier match and comparison. 

The feature vectors are used as the input of AI techniques for fault recognition. The step of fault 
recognition amounts to mapping the information obtained in the feature space to machine faults 
in the fault space. Numerous AI tools or techniques have been used classifiers and statistical 
learning methods have been widely used in fault diagnosis of rotating machinery, that includes, 
k-nearest neighbor (k-NN) algorithms , Bayesian classifier , support vector machine (SVM) and 
artificial neural network.   

2 
 

Theoretical background of AI 

Among the various AI algorithms 

k-NN 

Naive Bayes  

SVM  

ANN algorithms have been applied most commonly in fault diagnosis. 

2.1.​ k-Nearest neighbour  

k-NN is an instance-based learning algorithm based on the principle that the instances within a 
dataset will generally exist in close proximity to other instances with similar properties.  

For a given training set of classified instances  

where xi is the feature vector of the unlabeled instance, yi is the label. 

For a training sample x; y, the k-NN algorithm searches for the k nearest instances to x based on 
a given distance metric. The neighbourhood containing these k instances is represented by Nx. 

There are three basic elements in the k-NN algorithm: 

the number of measured instances k, 

the distance metric and  

the decision rule for classification.  

Compared with other AI algorithms, k-NN shows an advantage of simple implementation. 

3 
 

2.2. Naive Bayes classifier 

Naive Bayes first learns the joint probability distribution X; Y of the input and output by 
the conditional probability distribution based on the conditional independence  

Then, based on the learnt model, the output label y with the biggest posterior probability for 
the given input x can be calculated via Bayes’ Theorem 

Naive Bayes classifier is a common method for classification because it is easy to implement with 
high efficiency 

2.3. Support vector machine  

SVM is a computational learning method for small sample classification.  

Algorithmically, SVM builds optimal separating hyperplane f[x] = 0 between data sets by solving a 
constrained quadratic optimization problem based on the structural risk minimization 

2.4. Artificial neural network  

ANN​ is believed to be the most commonly used algorithm. For its most popular form,  

There are three components in an ANN: 

1. input layer,  

2.hidden layer and  

3.output layer. 

Units in a hidden layer are called hidden units, because their values are not observed. ANN is an 
intelligence technique based on a number of simple processors or neurons,The circles labeled 
‘‘+1” are intercept terms and are called bias units.r. An ANN interconnects many ‘‘neurons” and 
the output of a neuron can be the input of another. The weights W are obtained by an iterative 
training procedure, based on known input–output patterns. 

4 
 

ANNs implement algorithms that attempt to achieve a neurological related performance, such as 
learning from experience, making generalizations from similar situations and judging states 
where poor results were achieved in the past. 

2.5. Deep learning 

For many tasks, it is difficult to know what features should be extracted to feed the AI algorithms. 
Aiming at learning feature hierarchies with features from higher levels of the hierarchy formed by 
the composition of lower level features , deep learning methods have the potential to overcome 
the mentioned deficiencies in current intelligent fault diagnosis methods 

4. Discussion, limitation and future trends 

Various AI techniques of feature extraction and pattern recognition, as well as their applications 
to fault diagnosis, have been discussed. It can be seen that they have been applied in a wide 
range of problems of rotating machinery fault diagnosis. To sum up, they have their own 
advantages and weaknesses 

k-NN is an instance-based learning algorithm that delays the induction or generalization process 
until classification is performed. Compared with k-means, it only computes the instance of the 
nearest points instead of global distance. Therefore, it requires less computation time during the 
training phase than eager-learning algorithms such as Bayes nets and ANN, but more 
computation time during the classification process.  

Naive Bayes algorithm is characterized by the explicit underlying probability model, differently 
from other AI techniques; it provides a probability that an instance belongs in each class, rather 
than simply a classification. 

5 
 

SVM has excellent performance in generalization, also with few training data and thanks to its 
appropriate nonlinear mapping using kernel functions, data from two or more categories can 
always be separated by a hyperplane . Thus it can produce high accuracy in classification tasks 
for rotating machinery fault diagnosis and condition monitoring.  

ANN is a computational model that mimics the human brain structure, which consists of simple 
processing elements connected in a complex layer structure which enables the model to 
approximate a complex non-linear function with multiinput and multi-output. The structure of ANN 
can be versatile and by changing its architecture, it can achieve good fault diagnosis 
performance in many rotating machinery applications. 

Deep learning provides an effective way to learn features automatically at multiple levels of 
abstraction, allowing to learn complex input-to-output functions directly from data, without 
depending on feature extractors, which can be of great benefit for industrial rotating machinery 
fault diagnosis.  

Generally, SVM, ANN and deep learning methods tend to perform better when dealing with 
multi-dimensions and continuous features; while k-NN and naive Bayes tend to perform better 
when dealing with discrete features [110]. On the other hand, k-NN and naive Bayes algorithms 
are all explainable with clear physical meaning, whereas SVM, ANN and deep learning methods 
have poor interpretability. 

6 
 

7 
 

Conclusion 
Fault diagnosis for rotating machinery is vital to reducing maintenance costs, operation downtime 
and safety hazards. In this paper, a number of AI techniques have been surveyed for rotating 
machinery diagnosis. k-NN-based, Naive Bayes-based, SVM-based, ANN-based and deep 
learning-based fault diagnosis for rotating machinery have been summarized both from the 
theoretical background and industrial application points of view. With the AI techniques 
becoming more and more mature, it is believed that AI techniques will continue to be attractive 
and powerful for rotating machinery fault diagnosis. 

8