AI REGIO AI4CNC - A Federated Learning System Platform Development for CNCs
The AI4CNC experiment is designed to develop a Federated Learning System Platform for CNC machines to estimate tool wear through AI models and secure data sharing. The main objective is to use data from TEKNOPAR's CNC machines and sensors to estimate tool wear, replace tools at the optimum time to minimize costs and potential defects and address privacy concerns by proposing a Federated Learning (FL) model for CNC tool wear trained and executed on the edge. The AI4CNC experiment leverages federated learning to enable collaborative learning onedge devices while keeping the training data on the device. This approach aims to preserve data confidentiality and ownership by training a model using data collected from multiple CNCs at the real site of the experiment and then distributing the trained model to the data sources for edge execution. The goals were achieved by successfully implementing the planned activities and adjusting the approach to utilize sensor data instead of tool wear measurements for training AI models.
The AI4CNC experiment aims to build a Federated Learning System for CNC machines to predict tool wear, using data from TEKNOPAR's machinery and sensors such as accelerometers and microphones. The core intention is to determine the optimal tool replacement time, cutting down costs and potential flaws. With a focus on privacy, the Federated Learning model is designed for CNC tool condition. AI4CNC capitalizes on federated learning to promote collaborative learning, ensuring the training data remains confidential. This strategy underscores data privacy and ownership, as it trains the model with data from various CNCs at the experiment site and then disperses the trained model to the original data sources for implementation. The objectives were met through effective implementation of the designated activities and modifying the strategy to rely on sensor data over direct tool wear measurements for AI training. Combining predictions from various AI models further boosts accuracy. With a remarkable accuracy of 96.27%, the AI model effectively predicts the tool condition. Data stays confidential through federated learning, and the ensemble approach enhances prediction precision.