![](https://congressarchive.cimne.com/csai_2023/assets/banner_csai_new.png)
Digital Twin – Improving the Performance of a Semi-Robotic Screwing Station
Please login to view abstract download link
TDK Lambda, producer of power supplies, as a part of improving productivity of their assembly process, decided to convert a manual screwing station to a semi-robotic one. The semi-robotic station is integrated with human operations to improve the product quality, which relies on the tightening torque of the screws attaching electronic components to heat sinks. The main idea in this research was to develop a Digital Twin (DT) that will aid in monitoring operations, evaluating, examining and ascertaining the usage of a semi-robotic station and to provide a means to predicting quality issues and exploring in what ways it will help improving performance of the station. This work presents research performed on the new screwing station. This process was previously performed manually and produced both QA rejects and failure at customers, such as under-tightened screws that lead to overheating or may get loose and shorten circuits or over-tightened screws that may crack parts. A key component to the success of the project was producing a DT. Despite not having access to the internal design of the station, data collected was still able to improve the ability to capture problems that were not systematically captured before, detect new problems that were not known before and update projections for future failures. The DT was used for predictive maintenance, and offline, for testing ideas and improving ideas and schemes. The DT was developed using Matlab SimScape Multibody. Four schemes were developed: improving the screwing process; alerting; defect identification and correction; and prediction. A workflow was developed for the station DT. It is indicative both of the usage of the DT and of the information that needs to be collected to teach the DT how to detect known problems and identify new problems to introduce into the identification process and the correction procedure. The program runs, and if no problem is detected by the DT, it passes to visual QC. If a problem is detected, it goes into a learning process to check for deviations in the measurements coming from the screwing process. These are added to the DB and will be detected earlier in the process by the DT. Problems detected by the DT are classified and the system reacts accordingly. The digital twin so far helped detecting screws with faulty length, harmed heads, improper setup, missing and extraneous parts, defected parts, and updated the prediction of future failure of the station's screwdriver.