Machine Vision Based Quality Control and Fault Detection in a Textile Dyeing Machine

Abstract: Fault detection systems come in a variety of formats and are used in many different types of machines and industries. They can be used to perform fast and accurate detection, classification and analysis. The need for user interaction can be decreased and by that the general level of automation can be increased. This project has been conducted together with B\&R Automation and Imogo. B&R specializes in knowledge and products for the automation industry, including programmable logical controllers, vision sensors and vision solutions. Imogo develops new, more environmental friendly textile dyeing machines. In these machines the current fault detection system require more manual work and in order to further automate the system, one possibility could be to use machine vision. The purpose of this thesis is to investigate whether a B&R vision system could be used for fault detection and quality control in Imogo’s textile dyeing machine. A literature review has been undertaken, where the general topics of fault detection and machine vision have been investigated, as well as a more specific review of different potential solutions for the problem at hand. Different tests, where all key parameters, such as sensor configurations, lighting and resolution, have been performed in order to evaluate the system. The vision sensor along with the program have been tested and evaluated on the real machine and the result shows both advantages and disadvantages. The program is based on comparisons of mean grayscale and grayscale deviation values between an acquired image and an image of a correctly dyed piece. The system performs well on static fabric and manages to detect faults of different degrees. The main focus of the system is to detect if a fault has a occurred or not, especially large faults, but the program can also provide some additional information about the cause or location of the fault. For further development, a number of tests and different configurations can be done. With more knowledge and more testing this detection method has the potential to be both robust and dynamic, while still being sensitive enough. The project contributes to the field of machine vision and fault detection. This application differs in many ways from the common use cases for machine vision, which perhaps shows how versatile vision systems are.