A data mining approach to reduce the number of maintenance visits in the medical domain

We present an approach to reduce the number of maintenance visits for medical equipment using predictive maintenance. The proposed strategy considers that repair recommendations for an ensemble of equipments close to each other can be combined to one maintenance visit. For that purpose two recommenders that are trained with different false positive rate limits are used. The more sensitive recommender, i.e. the one with a higher false positive rate, is used to create repair recommendations that are only considered positive if a maintenance worker is already on-site or nearby. In case the travel cost is higher than the costs for the components to be replaced in the medical equipment, it is shown that a greedy recommender is helpful. As greedy recommender we consider an algorithm that recommends a replacement very early, which is further specified in the paper. Benchmark results suggest that this approach can actually reduce the total number of maintenance visits for the price that more components are replaced.