A Review of Artificial Intelligence Methods for Condition Monitoring and Fault Diagnosis of Rolling Element Bearings for Induction Motor

Table 3

AI algorithms used in CM and FDD of IM.

The method	Highlights
Random forest	(i) The small number of training samples is required; (ii) There is low computational cost; (iii) There is good performance for high-dimensional data
Bayesian network	(i) There is high classification speed; (ii) It is useful if the prior knowledge is reliable; (iii) There is low storage need; (iv) It is computationally expensive; (v) There is prior beliefs’ problem
KNN	(i) There is low classification speed; (ii) It is simple and easy to apply; (iii) There is poor performance for high-dimensional data; (iii) It is memory-intensive; (iv) It is noise sensitive; (v) It is computationally expensive
SVM	(i) There is good performance for high-dimensional data; (ii) There are low storage needs; (iii) There is high classification speed; (iv) It is not efficient for big data; (v) It is noise sensitive; (vi) It has good accuracy
ANN	(i) There is fault tolerance; (ii) There is high classification speed; (iii) There is parallelism; (iv) There is hidden training problem; (v) It is efficient for big data; (vi) It is computationally expensive; (vii) There is black box behavior problem
Neuro-fuzzy	(i) There is good performance for high-dimensional data; (ii) It has good diagnosis accuracy; (iii) There is robustness; (iv) There is parallelism; (v) It is efficient for big data; (vi) There is black box behavior problem; (vii) It has self-learning capability
DNN	(i) There is good classification speed; (ii) There are automatic fault diagnosis and detection; (iii) There is good accuracy; (iv) There is parallelism; (v) It has complex and deep architecture; (vi) It is feature extraction free; (vii) It is computationally expensive; (viii) There are massive parallel computations; (ix) It is efficient for big data; (x) There is long time training problem; (xi) A large number of training samples are required