Multi-task Bayesian compressive sensing for vibration signals in diesel engine health monitoring

•The learned dictionary is introduced for diesel engine vibration signals.•The reconstruction model is established based on the learned dictionary.•The Euclidean distance for measurement vector is proposed for classification.•The objective function is established to classify tasks into different clusters.•The measured signals validate the effectiveness of the proposed method. In the diesel engine health monitoring system, massive vibration signals are acquired and transmitted to the system center for failure detection and identification. To improve the efficiency of data transmission, an improved compressive sensing (CS) algorithm is developed for vibration signal processing by taking advantages of the multi-task Bayesian compressive sensing (MT-BCS) and classification theory. In this paper, the vibration signals are divided into different ‘tasks’ by rectangular window function to reduce the complexity of CS. with the effects of noises considered, the reconstruction model for the vibration signals is established based on the learned dictionary. Whereafter, Fuzzy C-means (FCM) is used to classify all the tasks into several clusters, which guarantees the existence of information sharing in each cluster. Thus multi-task Bayesian regression algorithm can be used for the tasks belonging to the same cluster to improve the reconstruction effect for diesel engine vibration signals. Finally, the effectiveness of the proposed multi-task Bayesian compressive sensing is validated by the experiments.