- As artificial intelligence (AI) becomes more and more important in industrial settings, the quality and consistency of the data fed into these AI systems is becoming crucial. The success of AI models heavily relies on good data quality. That’s why this thesis introduces a new system for monitoring the data going into AI models as it happens, this system is tested with historical data to make sure it works well. In industries where machines and production lines need to run efficiently and reliably, having high-quality data is a big challenge. Sometimes, the data quality changes or isn’t good enough, which can make the AI models give wrong results. This can make people lose trust in how well AI models work. Our paper tackles this issue by using a mix of methods - statistical, clusters and classes - to check the data in real time. We apply this to the data from an AI model designed to predict when cutting tools for glass manufacturing need to be replaced. We rate each checking method andAs artificial intelligence (AI) becomes more and more important in industrial settings, the quality and consistency of the data fed into these AI systems is becoming crucial. The success of AI models heavily relies on good data quality. That’s why this thesis introduces a new system for monitoring the data going into AI models as it happens, this system is tested with historical data to make sure it works well. In industries where machines and production lines need to run efficiently and reliably, having high-quality data is a big challenge. Sometimes, the data quality changes or isn’t good enough, which can make the AI models give wrong results. This can make people lose trust in how well AI models work. Our paper tackles this issue by using a mix of methods - statistical, clusters and classes - to check the data in real time. We apply this to the data from an AI model designed to predict when cutting tools for glass manufacturing need to be replaced. We rate each checking method and come up with an overall score. This way, we can accurately and efficiently evaluate both the data we already know about and new, unseen data. This overall score even helps someone who’s not an AI expert to quickly figure out if the AI model they’re using can be trusted right now, and it also points out when something might be off with the data. Looking ahead, we plan to fine-tune how we balance the importance of each method based on different situations. This will help make our monitoring system work well for all kinds of data going into AI models, not just in glass manufacturing.…
