Table 14.

Advantages and limitations of the hybrid models employed for mechanical fault detection and fault prognosis

Publication	Advantages	Limitations
[45]	Proposed approach: detects and classifies different mechanical faults in unlabeled data.	Not identified
[50]	LSTM: captures nonlinear relationships in sequential data.
	ARIMA: models linear associations present in time series data.	Not identified
	Proposed approach: optimization of the performance of the proposed fault prognosis model.
[59]	WPGMC clustering: interpretability.
	RNN: captures complex, nonlinear relationships in time series data.	Not identified
	Proposed approach: uncovers patterns of wear and tear in unlabeled data.
[74]	Random Forest: improved performance; robustness when handling numerical data and real-world problems.	Not identified
	Proposed approach: improved accuracy.
[58]	Autoencoder: can learn the relationship between the input data variables.	Proposed approach: anomaly threshold is defined arbitrarily; could use more a more sophisticated model to improve prediction accuracy.
	Proposed approach: can learn from unlabeled data; applicable to different domains.
[75]	GMM: capable of reducing the number of clusters.	Not identified
	FP-Growth: handles large databases efficiently; can handle itemsets with low support threshold.
	Proposed approach: interpretability; can handle different types of sensor data; simple to set-up.