<p>To implement the concept of Predictive Maintenance (PdM) effectively in Industry 4.0 with Big Data, the Internet of Things (IoT) and machine learning are used to incorporate various condition monitoring data as well as faced the limitations while predicting the Remaining Useful Lifetime (RLU) components. On the other hand, the annotation of the component is considered the most difficult task for identification and thus may lead to the issue in attaining quality labeled data. Significantly minimizing cost production and guarantee increased uptime throughout the prediction chain are the two major objectives for the Industry 4.0 applications in PdM. This research work is aimed at developing an efficient PdM system in Industry 4.0 using a deep learning mechanism. Initially, the required data is collected from the sensor devices, which are connected to the smart manufacturing system. The data cleaning process is carried out and then, the stacked and the spatial features are obtained using Long Short Term Memory (LSTM) and the 1-Dimensional Convolutional Neural Network (1DCNN). Consequently, the target-based feature pool is created and then fed to the Parameter Tuning-Dilated Recurrent Neural Network (PT-DRNN)-based prediction system to predict the failures in the industrial systems. Here, the parameters are optimized using a developed Hybrid Artificial Flora Snake Optimizer (HAFSO) to enhance the prediction performance. Finally, the test results are verified by comparing the performance measures over the currently developed PdM systems in Industry 4.0. Our proposed model HAFSO-PT-DRNN reached a higher accuracy rate of 98.29% while the traditional models such as DHOA-PT-DRNN, WOA-PT-DRNN, SO-PT-DRNN; AFOA-PT-DRNN attain an accuracy rate of 91.5, 91.09, 94.02 and 93.67%, respectively on the prediction performance when analyzing with Dataset 1.</p>