Machine learning enhances the production of a smart factory
”Good that we didn’t tell you that this is impossible. Great results!”
Philip O’Leary, Data Integration Manager, Murata Electronics
Manufacturing semiconductor components from raw materials to finished products is a long process that requires extreme automation precision, cutting-edge technology, and almost perfect cleanliness of the workspaces.
Dozens of manufacturing equipment in different stages of production lines store vast amounts of measurement data, which tells about the manufacturing stages’ progress and operation of the equipment. Murata uses data to improve the manufacturing process yield and identify error situations and the root causes of possible problems.
Murata is a globally leading manufacturer of electronics components and solutions. The company develops and manufactures MEMS sensors, which measure, for example, acceleration, tilt, vibration, and pressure.
Murata’s products are used in applications that require extreme reliability and precision. Thanks to the solutions, e.g., cars stay in their lanes, bridges don’t collapse, and hearts beat in people’s chests.
ATR Soft is Murata’s long-term partner in developing and maintaining IT systems. In addition to the company’s integration platform and BI solutions, in recent years, we have deepened our cooperation also to cover the company’s data processing needs. We are developing the analysis of data collected in manufacturing processes using artificial intelligence, in practice, machine learning, with the goal of an even smarter factory.
With the help of modern machine learning methods and advanced program libraries, we have developed models from the raw data collected from production, which, e.g. recognize the device’s problem in advance and perform quality control with better accuracy than humans. In many applications, the amount of data to be interpreted and classified using machine learning is so large that it would be impossible for a human.
The Python environment we use is a natural choice for data processing and machine learning. Many solutions are based on the PyTorch machine learning framework.
The Fast.ai deep learning library facilitates the developer’s work and offers good tools for teaching the model and interpreting and explaining the results.
With time series data, the tsai library enables using the latest model architectures through a familiar programming interface Fast.ai.