Table 12.

Vibration FD methods used in belt conveyor idlers via ML models.

Authors	Detection Methods	Accuracy	Main Findings	Advantages	Disadvantages
Li et al. [4]	WPD and SVM techniques	100%	The results of an experiment performed on a belt conveyor in a coal mine demonstrate that the proposed system can find faulty idlers with limited sensors.	An SVM classifier can correctly identify normal conditions and several faulty conditions.	The computation time of the classification results in a delay in the detection of faults.
Muralidharana et al. [58]	Statistical features and decision tree algorithm	99.52%.	The results of this study may not be generalizable to all cases. Nevertheless, the methodology employed will serve as a guide for future research in this area.	Simple to understand and interpret for feature selection and classification.	The decision tree provides more information about the classification of faults. It can result in trees that are overly complex and are not able to generalize the data well.
Ravikumar et al. [46]	Statistical features and Artificial neural network and naïve Bayes (NB) algorithm	ANN: 90% NB: 95%	Results from artificial intelligence and NB demonstrate the accuracy of the NB performs better than the ANN algorithm in predicting self-aligning conveyor roller failures and assessing their lifetime.	NB performs better due to its simplicity and assumption of feature independence.	Since the number of faults collected is limited, the proposed method cannot be generalized.
Bortnowski et al. [44]	LSTM Autoencoder model	90%	The use of the autoencoder facilitated the automation of damage detection, which is invaluable when assessing the operation of long-distance conveyor routes.	The proposed method was able to detect the location of potential roller damage based on the change in the average peak frequency over time and spectral autocorrelation.	Low-frequency signals related to the specific operating conditions were regarded as potential sources of damage when they were simply sources of signal interference.
Roos and Heyns [23]	Wavelet package decomposition and artificial intelligence	100%	The use of SVMs for vibration monitoring in-belt systems for conveyor idler bearings is strongly recommended, together with WPDs for preprocessing the signals from the bearings.	SVM can identify bearings that are in the early stages of failure, even with added payloads, compared to ANN.	A certain kernel function may be more accurate in classifying some data sets in SVM. The impact of payloads on signal clarity affected the accuracy of the proposed methods.
Ravikumar et al. [45]	Statistical features and k-star algorithm	91.7%	It is found that statistical features and K-star algorithms are effective tools for detecting faults in self-aligning troughing rollers on bulk material handling belt conveyors.	Since k star is an entropy-based algorithm, it can handle a wide range of complex data sets better than ANN.	A new instance of a class is assigned by comparing the closest existing instance with the new one using a distance metric, which is inefficient in terms of memory usage.
Ravikumar et al. [48]	Statistical features and random forest algorithm	90.2%	Random forest results summarize the accuracy of the algorithm in terms of predicting self-aligning conveyor roller failures as well as assessing the lifetime of the conveyor rollers.	The accuracy of Random Forest is generally high.	Random forests can be computationally intensive depending on the data collected.
Ravikumar et al. [47]	Statistical Features and Support Vector Machine	98.08%	According to SVM results, the algorithm accurately predicted self-aligning conveyor roller failures and assessed conveyor roller life expectancy.	An advantage of the SVM Algorithm is that it can handle multiclassification tasks.	It can be difficult to select the appropriate Kernel functions.
Wijaya et al. [60]	Wavelet transform and artificial neural network	99%	By adopting the proposed fault detection scheme, fault identification and classification can be accomplished accurately and unaffected by changes in operating modes, such as conveyor belt speed.	Three idler conditions tested using ANN provided more than 99% classification accuracy even with varying belt speeds.	Due to the limited number of faults and the inability to simulate the effect of loading on the conveyor frequency signatures, the proposed method was untrustworthy.
Wijaya et al. [12]	A high-frequency energy and envelope spectrum and IForest method are used to determine fault location	90%	According to the study, faulty idlers significantly increased high-frequency energy with a bearing fault detected through envelope analysis.	Most anomalous data points were detected with a 90% reduction in analysis time.	The disadvantage of this method is that data were directly divided into half rather than randomly determining where to split the data.