At MAZIN, Inc., we are dedicated to developing AI solutions to address production challenges at the process level. In the field of machining, we have focused on creating algorithms that monitor tool conditions based on electrical current values, which correlate with cutting torque. This initiative involves clamping current sensors to machine tools and analyzing the data to detect any abnormalities.
Previous iterations of MAZIN, Inc.'s tool anomaly detection algorithms were tailored for mass production processes and did not cater to small batches or diverse product lines, including one-off items.
However, when machining large, unique workpieces in small batches, tool anomalies can occur mid-process. If these anomalies go undetected and machining continues, it could result in unintended machining and turn the workpiece into scrap.
This issue is particularly critical for large workpieces due to their high value, making the challenge of tool anomaly-induced defects more pronounced.
To address these challenges, MAZIN, Inc. has initiated the development of a tool anomaly detection algorithm suitable for gantry-type machining centers that process large workpieces.
The algorithm is designed to target operations that involve periodic machining of the workpiece.
When machining small batches of diverse products, targeting periodic waveforms for the algorithm can mistakenly identify non-periodic waveforms as tool anomalies. We have developed an algorithm that overcomes this challenge.
For anomaly detection, we used test data that introduced slight increases in current values within periodic waveforms to create anomalous sections. Analysis of this data revealed distinct two-dimensional distributions for normal and anomalous conditions.
By inputting data obtained from the beginning of the machining process into the anomaly detection algorithm, it was observed that the anomaly degree values on the vertical axis significantly increased during the anomalous sections, as illustrated below.
Notably, the initial part of the waveform, representing the non-machining section, also showed different values from normal conditions. However, the anomaly detection algorithm correctly identified only the anomalous sections as abnormal.
This test confirmed that our anomaly detection algorithm could accurately identify anomalies within specific sections. Moving forward, we will verify the algorithm's effectiveness with actual waveforms generated during machining anomalies in real production settings.
For more details on the technology or inquiries about our research and development, please contact us.
