Unexpected machine breakdowns cause production losses, maintenance costs, and even safety issues in contemporary industrial settings. In order to realize reliable and efficient operation in this regard, it is crucial to predict machine failure in an accurate and timely manner. Traditional predictive maintenance solutions tend to solely depend upon information acquired through singular sensors or rules for monitoring this issue. With this perspective in mind, this study proposes a multimodal deep learning method that combines several sources of industrial data for estimation regarding machine failure risk. The architecture brings together the vibration signals, acoustic, temperature, and processing logs to create an integrated view of the state of the machine. The application uses Convolutional Neural Networks for spatial features and BI-LSTM or RNN layers to capture the temporal relationships that exist in the sequential dataset. The attention fusion module enables the system to focus on the most relevant sensor channels for the specific failure mode. Experiments show that the system performs better in terms of prediction accuracy, early fault detection, and the frequency of false alarms than state-of-the-art solutions, making it ideal for the Industry 4.0 framework in smart manufacturing.
Nagaraju et al. (Thu,) studied this question.