Analytical Root-cause Identification in data Streams for detection of Emerging quality issues

Project start:
1 September 2016
Project end:
31 December 2018
More info (PDF):
[[media: | pdf]]
Slawomir Nowaczyk
Application Area:
Intelligent Vehicles

Involved internal personnel
Involved external personnel
Involved partners


FFI Vinnova project

One of the top priorities for commercial vehicle manufacturers is product quality. At the same time, vehicles are becoming more and more specialized, due to customers demanding higher productivity, and this vehicle diversity makes it harder to predict and detect quality problems. Telematics solutions today enable OEMs to monitor their vehicles while in operation and have been successfully used for predictive maintenance purposes, but the work to use similar data to enable earlier and more precise detection of upcoming quality problems is just starting.

Significant improvements in this area will only be made possible by combining on-board data with existing in-office knowledge such as warranty claims, technical reports and expert knowledge. Quality problems can be detected earlier by identify emerging patterns, discovering trends, and detecting anomalies. Better understanding of the issues will also allow for more precise solutions to be applied, for example by choosing between vehicle recalls, redesigning or updates to usage guidelines.

Arise will develop algorithms for early detection of quality issues and their analysis, integrating multiple available data sources. We will provide quality analysts with data mining and machine learning methods capable of extracting patterns and finding trends in these diverse data sources, using methods such as transfer learning and causal relation discovery. This will allow them to spot quality issues significantly sooner than today and to identify their causes with significantly higher accuracy. The essential step towards this goal, and the first result expected from the project, is development of methods that aid in identifying the right root cause and categorizing new warranty cases as they are entered into the warranty system.

The solutions developed in this project will allow for automatic and semi-automatic highlighting and prioritising of the most relevant issues, based on online and incremental algorithms especially suited for early detection of problems in the product launch phases. The importance of this work is clear, since a recall campaign can cost millions of SEK.

To summarize, the main research question for the project is: can a semi-automatic method find meaningful and understandable patterns in the warranty-related data, and can those patterns improve future reporting, detection and containment of quality issues? The project aims to answer this question in a way that combines the two aspects mentioned above, thus leading to a significant decrease of costs and increase of customer satisfaction for the Volvo Trucks.

Faster and more accurate quality control increases the overall product development efficiency. This strengthens competitiveness of Volvo as well as the Swedish automotive cluster. Solutions developed in Arise pushes the technology forefront of Big Data within the automotive industry. This will increase the need for, and importance of, highly skilled personnel and providing new career opportunities in Sweden. Safer and more environmentally friendly vehicles with higher quality reduce waste and contributing to a more sustainable society through a prolonged product life time. Arise will increase the knowledge and competence on data analytics within Volvo. With more and more data being collected by the automotive industry, it is important to continuously explore new ways to use this information for increasing competitive edge. This will strengthen innovative capacity, not only within Volvo but also in Sweden. Arise will contribute to the Swedish research community through the cooperation between Halmstad University and Volvo. The research will focus on machine learning algorithms for doing pattern recognition and statistical modelling on streaming data.

The project will showcase and quantify the business benefits that are associated with Big Data. Unknown or wrongly categorized quality problems can be very costly and any method that allows for earlier and more precise detection is very valuable for vehicle manufactures. Arise will demonstrate that by combining multiple already available data sources it is possible to reduce the detection time and increase the root cause precision, which in return lowers the impact of quality problems.

The first result from this project will be the development of online and incremental algorithms for decision support based for early detection of quality issues (especially, but not limited to, the launch phases of new vehicle models). Those algorithms will be looking at streams of data from multiple sources (e.g. warranty claims, vehicle usage statistics, configurations and maintenance). By processing information as soon as it arrives, we will be able to give initial warnings immediately, even based on very few problem reports.

Another important results from the project will be tools and methods for supporting workshop mechanics who report new warranty claims. They can greatly impact the process of detecting and containing quality issues, but they often lack the expertise needed to make good decisions in this regard. Additionally, at the moment they do not receive the support they need. In particular, a more global perspective on the quality would help those mechanics to create more useful warranty reports, with better problem descriptions and more relevant background and context information.

Finally, when a quality issue is identified, the most important task is to understand it well enough to decide the best way to contain it. The first step towards that goal is figuring out which vehicles are affected – what are the common traits that can be indicators, or causes, for this issue. Precision of this step is going to have huge implications in how costly and how successful any solution will be.