<p>Fault detection and prediction plays a significant role in the chemical process industries to maintain safety, reduce the downtime and increase the operational efficiency. The chemical, petroleum and pharmaceutical plants are industrial systems that have complicated and delicate process and any faults that are not detected may have serious safety, environmental and economic repercussions. Tennessee Eastman Process is a popular benchmark simulation, modeling the conditions of the real industry, a number of fault scenarios, measured variables, and control parameters. Nonlinear, high-dimensional, and time-dependent however forms a challenging problem of analysis to the conventional machine learning methods in fault analysis. Despite the good performance of the available methods like LSTM, Transformer-based model, CNN-LSTM, and graph-based methods, they are usually associated with low interpretability, inefficient extraction of time features, and unstable predictive ability. This paper will solve these shortcomings by suggesting a superior explainable deep learning model of multi-fault detection and prediction. The structure combines the Temporal Convolutional Networks to extract temporal features, the Probabilistic Neural Networks to identify faults and the Bidirectional LSTM to predict the Remaining Useful Life. Further, the explainable AI using SHAP is used to explain the contribution of features and improve the transparency of the model. As shown in the experimental findings, the proposed approach can be used to conduct reliable predictive maintenance in multifaceted industrial settings because experimental results show better accuracy, robustness, and interpretability of the approach as compared to the current ones.</p>