| dc.description.abstract | In collaboration with ENRX, this project conducted a technical feasibility study focused on data collection and analysis. The project involved gathering operational data from induction hardening machines owned by various company clients. This was collected by designing and installing a data-logging system on machines belonging to three of the clients. The collected data went through an initial phase of exploratory data analysis before being utilized to construct an LSTM (Long Short-Term Memory) Neural Network for error prediction.
During the first part of the project, the collected data sets were examined to reveal both linear and non-linear correlations. This analytical study provided valuable insights into how clients use the different machine parameters. It also identified variables that showed strong correlations, those that were less relevant to the specific case under investigation, and potentially missing variables that could be of interest.
The second phase of the project involved the creation of two LSTM models to predict a particular machine alarm. In the first case, the objective was to forecast the specified alarm by analysing the initial 10 seconds of a hardening process. The second case analysed the preceding five hardening processes to determine whether the next hardening process would encounter the targeted alarm. For both scenarios, the data required for input into the LSTM models was accordingly pre-processed. After evaluating the model results, it was found that the second study case performed better than the first one, showing the value of the data collected from clients. These findings highlight the project’s potential for further development and its capacity to unlock promising possibilities in the field. | |
