Performance Comparison of User-Estimated and Learned Task Recognition Models for Maintenance Procedures
Behrendt, Marcus and Wortelen, Bertram
HCI International 2021 - Late Breaking Papers: HCI Applications in Health, Transport, and Industry
Maintenance procedures of technical installations are crucial to keep systems in a healthy state. Unfortunately, sometimes rare technical installations require certain types of maintenance that are not known to all technicians. We are investigating an idea that enables those technicians to still perform maintenance by using an assistance system (AS). An important part of such an AS, which we concentrate on in this work, is a task recognition engine. It keeps track of the tasks performed within a maintenance procedure, so that the AS can provide the technician with adequate information and help. We show how to build two types of task recognition models for this task recognition engine. The first was developed by Honecker & Schulte and is based on the Dempster-Shafer theory (DST). These models can be well estimated by experts, but may not represent the sequence of tasks in maintenance procedures because they are designed for parallel tasks. But still, they may be useful if there is not enough training data available. The second is based on Hidden Markov models (HMM). They are harder to estimate, but they may perform better for the sequential tasks in maintenance procedures if they are trained from a sufficient data set. Depending on the training data set's size, we investigate which kind of model performs better by evaluating them on an example maintenance procedure.
2021
inproceedings
Springer International Publishing
405-426
