Using audio to detect errors in manufacturing
Since the rise of industrialism, factory manufacturing has been an important part of our society and its economy. In factories, machines produce goods on assembly lines, while human operators either monitor the process, or control the machines themselves. The way each machine is built differs, and thereby also the way deviations in the process are best recorded. Artificial Intelligence can be used in different ways when analyzing deviations, where sound can be an important aspect to analyze.
In the sub-processes of machines where hearing is an important observation tool, AI can be an extra pair of ears that either gives confirmation of something the machine operator has already observed, or warns of errors that the operator himself failed to detect. This could contribute to increased social sustainability of operators with impaired hearing due to age or injuries. Inexperienced operators can also receive support in their decision-making, which could provide increased security in the work. In addition, there is an opportunity to reduce boring work tasks if AI can completely take over the task, which can contribute to improved health as studies have shown that chronic boredom at work can cause depression, anxiety, stress and insomnia.
From a security perspective, AI can also detect small errors before they become large. A small error is often easy to correct, while larger errors, especially in larger machines, can result in production stopping and a worker having to enter the machine itself to correct the error. Even with safety regulations, these work steps can involve risks for workers, which could have been avoided if the error had been detected earlier.
The following project took place based on a course in sustainable development and ethics in computer technology, in collaboration between Gothenburg Technical College and students at Chalmers University of Technology. The project's goal was to construct a prototype for a product that analyzes incoming sounds in real time and then determines whether a deviation has occurred or not. The prototype has been constructed in two parts, where one part was implemented through machine learning and another part was implemented through a more classical analysis of frequency spectrum. The two methods have been compared and an analysis based on sustainability and ethics has been carried out, focusing on the impact the prototype could conceivably have on workers, companies and society.
The complete report is in swedish and can be found here.
The summary and report are written by Agnes Brogeby, Victor Ebbesson, Elliot Hultgren, Andreas Nilsson, Sebastian Rizk Gustavsson, Max Vallin Ek Datascience students from Chalmers.
Tags
This article is tagged with these tags. Click a tag to see all the articles with this tag.