Applying a Data Analytics Approach to Medical Equipment Maintenance Management to Improve the Mean Time Between Failure and Availability

Abstract

Maintenance of medical devices has become a top priority around the world, this is due to the vital role of these devices in contributing to saving human lives. Maintenance engineering makes use of data analytics using statistical techniques to expose patterns to capture relationships among the different variables that affect the performance of medical devices. This research presents an attempt to discuss an efficient approach of data analytics for maintaining medical devices to enhance their reliability in terms of the mean time between failure and availability. Data analytics methodology is applied based on supervised machine learning techniques by training the classification models to improve reliability, availability. Raw data were collected from the cooling system of the Magnetic Resonance Imaging MRI magnet as a case study, then organized and cleaned to make it suitable to build and train classification models. A new representation of the data has been resampled using the Synthetic Minority Oversampling Technique (SMOTE). Six different classification models were trained and tested using Weka software, these are Logistic regression, K-nearest neighbors, J48 and Random Forest decision trees models, Support-vector machine SVM, and Naive Bayes. Based on the F-measure, K-nearest neighbor, Random Forest, and SVM were chosen to be the most significant classifiers, a comparison between these three classifiers was done based on evaluation metrics, the metrics are: Correctly classified instances, Incorrectly classified instances, Root mean squared error (RMSE), Recall, Precision, F-Measure, Area under ROC curve, and Accuracy, it was found that the random forest classifier has the best performance, highest accuracy, and the least error.Maintenance performance in terms of MTBF and availability are estimated after applying the random forest model and compared with their values before to determine the efficiency of the classifier. It was found that the random forest model has caused an improvement in MTBF by 879%, and an improvement in the device availability by 11%. Data analytics plays a critical role in improving efficiency and generating understandings from the data generated by devices. Through this paper, data analytics was followed for maintenance planning to improve the mean time between failure and availability of the cooling system of the magnetic resonance imaging device, the results were positive and encouraging. The same methodology can be generalized to benefit from it in maintenance planning for other medical purposes.