NordDesign 2016 August 10 – 12, 2016 Trondheim, Norway Towards big-data analysis of deviation and error reports in product development projects Ívar Örn, Arnarsson 1 , Johan Malmqvist 1 , Emil Gustavsson 2 , Mats Jirstrand 2 1 Product and Production Development, Chalmers University of Technology varo@chalmers.se, johan.malmqvist@chalmers.se 2 Fraunhofer Chalmers Centre (FCC) for Industrial Mathematics, Chalmers Science Park emil.gustavsson@fcc.chalmers.se, mats.jirstrand@fcc.chalmers.se Abstract Large complex system development projects, such as complete truck development projects, take several years to carry out. They involve hundreds of engineers who develop tens of thousands of parts and millions of lines of codes. During a project, many design decisions often need to be changed due to emergence of new information. The bulk of these changes are requested late in the development process. It is known that changes late in the development process are very costly and run a risk of delaying the project. These changes are often well documented in databases, but, due to the complexity of the data, few companies analyze engineering change in a comprehensive and structured fashion. This paper argues that “big data” (specifically data mining) analysis tools can be applied for such analyses and proposes a process for carrying out the analysis and using the results for product and development process improvement. The paper further accounts for experiences gained from testing the approach on a dataset consisting of 4,000 deviation and error reports that were created during a truck development project. Keywords: Product development, late product changes, big-data analysis. 1 Introduction Truck development projects involve hundreds of engineers developing tens of thousands of parts and millions of lines of codes. The projects have durations of several years. During the course of a project, many design decisions are made. Many decisions need to be changed during a project due to, e.g., the emergence of new information, simulation or test results. Unfortunately, the bulk of change needs tend to be discovered late in the process, as shown in Figure 1 (Giffin et al., 2009).