AI-based tinting assistant for batch optimisation

 
Tinting or shade matching is a labor-intensive and time-consuming process, particularly in industries such as automotive and architecture, where achieving precise colors is crucial. This requires validating the color of produced paint by applying it to a new panel, drying the panel, comparing it with the standard panel, and adding the tinters to the base paint for color adjustment to achieve the target color. The selection of tinters and their weights is determined by the tinting expert based on his/her past experience on similar batches. It requires several iterations to achieve the target color due to variations in base products, tinters, and several other factors. To streamline this, we did co-innovation with an automotive paint producer to standardize and optimize the tinting process through liquid color measurement combined with an AI-based tinting assistant. Initially, we conducted research to identify a suitable color sensing technology in the industry. A specific mechanical structure was created to mount the sensing technology for providing reliable and repeatable measurements of the liquid and panel colors. A new digitization platform was developed to digitize the data of various batches, products and tinters through an operator interface. A system with AI assistant was developed to capture the tinting expert's knowledge and recommend the tinters with dosing quantities once enough knowledge was acquired. The AI assistant required data of various products and tinters to accomodate different shades of colors during the learning phase. During the validation phase, the accuracy of AI model was determined through comparison of the predicted tinters with the actual tinters dosed by the operator. Initial model accuracy was low however, it increased substantially with training on more data for continuous 3 months. Wit more analysis, it was observed that the model provided different accuracy for different shades of colors. The AI assistant system provided shortest path to reach the target color depending on the base color, tinter strength, and other parameters that were used to train the model. The system provided an interface to the operator to override the predictions. Furthermore, the system was provisioned with self-learning capabilities to improve the future predictions by learning from the operator actions and the errors observed from the subsequent measurements. This approach can minimize the number of iterations in tinting process and help the tinting operator finish more batches in the same time. Liquid color measurement would reduce the number of panels created per batch and thus achieve more sustainable quality control. The digitization of the tinting process would help to provide better insights to the stakeholders such as the plant head and quality head to gain better predictability of the production. Ultimately, the solution would help to achieve faster time-to-market, increased OEE, reduced energy consumption and lower production cost.