30 credits - Using historical vehicle data with deep learning for predicting health in commercial vehicles i Sodertalje

In the coming years the transport system for goods and people will undergo significant changes. Upcoming technological changes will reshape the value network and have a significant impact on future business models. Scania is undergoing a transformation from being a supplier of trucks, buses and engines to a supplier of complete and sustainable transport solutions. A critical success factor in the shift to becoming a world-leading provider of transport solutions is operational availability of vehicles. The belief is that the importance of uptime will be even further accentuated going forward.

Background

Being able to use vehicles according to plan and without the risk of unplanned breakdowns is fundamental to ensure an efficient transport system. If a vehicle’s health status can be accurately predicted or forecasted there is the potential to streamline maintenance, increase operational availability, and reduce the risks of costly repairs and destroyed cargo.

Accurate health status predictions are fundamental to ensure operational availability. However, because of the diversity in the way trucks and buses typically are configured and used, accurate prediction models are complicated to achieve. In addition, data-driven prognostics rely on large amounts of historical failure data to estimate prognostics model parameters and this type of data is limited in real-world industrial scenarios. This makes it difficult for data-driven models to extract degradation patterns and characterise system performance from historical data.

Data-driven prognostics models for component health status typically utilise vehicle, customer, and vehicle usage data. However, another source of information that has been far less explored in the domain is vehicle service history. In medical science, electronic health records have been utilised with success to predict primary diagnosis category for a patient’s next hospital visit given previous visits. Since vehicle service history data to several aspects is similar to electronic health records there are reasons to believe that developed methods for medical claims could be translated to the commercial vehicle domain.

In recent years various deep learning models have been applied to the predictive modelling of medical claims. In these studies each patient is represented as a sequence of temporally ordered irregularly sampled visits to health providers. Since different patient conditions have different temporal progression patterns successful models have a mechanism that learns time decay factors for every medical condition. The same type of temporal progression patterns are expected to be observed for vehicle malfunctions.

Assignment

The project will develop methods and knowledge for how to integrate vehicle repair data in data-driven prognostics models for component health status. The work is intended to create and train a recurrent neural network model with time decay mechanism for preliminary investigations of service history based prognostics. The methods will be evaluated for one uptime critical component.

Education and skills

Master’s student in computer science, mathematics, physics or similar, preferably with specialization in statistics, machine learning, artificial intelligence and data science. Documented experience and skills in Python in addition to machine learning and deep learning, is a merit.