THE CHALLENGE

ARTiTEX is a start-up from Bielefeld and wants to revolutionize the textile industry. It relies on smartphone technology and our skill in artificial intelligence.

What does AI have to do with the textile industry? Nothing at first – But: ARTiTEX software provides users with information about when, which worker, made what, which provides an incredibly valuable overview of the processes. However, this comes with a catch, and that is unfortunately this information does not fall from the sky, meaning each worker must tap a button in the app each time a process is completed to report the completed process. This is unsafe and costs both time and nerves. We then proposed an automated solution in the form of an AI.

To do this, the worker places their smartphone on their sewing workstation and then records some sewing processes at the touch of a button. Audio and vibration signals are stored so that we, as experts for the skillful handling of data, can link the AI to it, which, after a short time, takes over the button pressing permanently, so to speak.

SOLUTIONS

The aim is to automatically recognize the separate sewing processes of each workstation separately, so that information is available in real time about which workstation has produced how many parts and when:

• The workers are each accompanied by an artificial intelligence in the app, which must first hear (audio) and sense (vibration) processes in order to learn from them.
• For this, a recording process is started/saved via touch function, so that after approx. 20 min or 100 processes, enough data is available to train the AI with it.
• After completing the training, the AI is now permanently available. It listens to and feels all the sewing processes of the worker in the course of the working day and counts them.
• No manual entries and no button  necessary anymore! Instead, all desired information is provided by the AI reliably, automatically and in real time.

Results

The following results were obtained:

• 100% accuracy in detecting the processes regardless of various background noises
• 87% probability that the time of detection of a process and the actual start time of the process have a deviation of less than +/- 30% of the process length.
• The AI generates the above results under real manufacturing conditions. In doing so, it successfully distinguishes similar processes if they have been performed at a different workstation, allowing isolated observation of any worker.