DISA

When? January 13th, 14-16
Where? D1140 – Växjö (link will provided for those who wants to attend online)
Registration: We would like to know how many that will attend onsite/online in order to get some fika for those onsite. So please register by January 11th https://forms.gle/mxHmRtdEydUWGoa79

Agenda
14.00-14.10 Welcome and practical information from Welf Löwe
14.10-14.55 Presentation and discussion: Exploiting Automatic Change Detection in Software Process Evolution for Organizational Learning – Sebastian Hönel
14.55 – 15.05 Coffee break
15.05 – 15.50 Presentation and discussion – Design and implementation of factory-integrated machine learning models and case studies of ongoing data-driven projects – Felix Viberg
15.50 -16.00 Sum up and plan for our next seminar on February 13th

Abstracts

Exploiting Automatic Change Detection in Software Process Evolution for Organizational Learning – Sebastian Hönel

Software project management is a multifarious activity, with the software development process being one of its facets. It was previously shown that there is a strong positive correlation between the quality of the process and the quality of the manufactured product. Measuring the quality of the software process, however, is difficult. This is especially true for phenomena impacting the quality which have manifestations not only in metrics but also in the project’s culture and management. Another hinder to efficient organizational learning is a scarcity of data and a consequential lack of generalizability.

There exist suites of metrics to quantify the development process. However, many metrics are difficult to use because they are not scores or do not relate directly to defined quality goals. While some metrics can measure certain properties, they act as a surrogate to the otherwise latent, more interesting quantity. The objectivity of the metrics captured, as well as minimizing the error, is of great importance for predictive models.

We suggest that a synthesis of objectively captured digital artifacts from application lifecycle management, together with qualitative analysis can be exploited for efficient organizational learning. We first propose a new, language-agnostic metric, which reduces noise in changesets. It can be used to infer the type of activities as they come along with typical software projects. Such activities are then used to capture the actual sought-after phenomena.

We can empirically prove that certain phenomena are well captured through objective artifacts only by this synthesis of data in some contexts. Choosing appropriate stability criteria ahead of time allows for obtaining robust regression models using adaptive training from “little data”. Such models have the power to make future qualitative analysis obsolete, which organizational learning has the potential to greatly benefit from.
For more information about Sebastian Hönel and his research see: https://lnu.se/en/research/research-projects/doctoral-project-efficient-detection-of-changes-in-software-evolution/

Design and implementation of factory-integrated machine learning models and case studies of ongoing data-driven projects – Felix Viberg

Even though the capabilities of modern machine learning-based solutions are thoroughly demonstrated within academia and elsewhere, such solutions are yet to become truly valuable within manufacturing. Some of the reasons for this might be (i) connectivity limitations between controllers and computers stemming from deficiencies within existing communication standards, (ii) a lack of fundamental capabilities within the factory network, or (iii) reluctance to invest in the heavily cross-organizational team needed to develop the solutions.

This talk aims to demonstrate design choices and implementations of support systems for accelerated machine learning deployment within traditional manufacturing settings, based on the three deficiencies stated above. Two ongoing projects, jointly serving as a platform for experimentation within production, is discussed and demonstrated up to their respective current state. The first project is an application for autonomous in-line visual inspection based on CycleGAN, and the other an attempt to use Normalizing Flows on high-frequency servo-drive data.

For more information about Felix and his research see: https://lnu.se/en/research/research-projects/doctoral-project-machine-learning-in-manufacturing/