The combination of Mitsubishi Electric's laser processing knowhow and AIST's machine-learning technology for image recognition enables automatic evaluations of edge quality, comparable to manual evaluations performed by experts. Various factors can lower processing quality, such as debris on the lens used to focus a laser beam, changes in the temperature of the machine and the condition of the workpiece surface. Conventionally, experts must check the cut surface to evaluate the resulting quality and then make adjustments as required to improve the quality.
The new automatic quality evaluation technology of cut-surface images achieves the level of manual evaluations performed by experts. In addition, a newly standardised procedure allows laser-processing machine operators to change settings, based on the automatic evaluations and thereby improve processing quality without need of expert skill.
In the case of robots, the new development is Mitsubishi Electric's force-feedback control technology and AIST's machine learning technology for data analysis, which are combined in a machine-learning system that classifies anomalies in the assembly operations of industrial robots. This reduces the time to create programs to cope with anomalies by some 66%.
In setting up industrial robot systems, a high volume of programs is required to cope with anomalies, which differ from system to system. Conventionally, experts predict possible anomalies in advance, such as errors in grasping or alignment, and then develop the many programs required to address the anomalies and recover normal operations. Developing such programs, however, takes far more time than the actual programming of normal robotic motions. The new technology learns to classify anomalies, using force-sensor outputs obtained during operation.
A third development is drive parameter tuning on machines requiring high speed positioning of gantry-style machinery, although the specific machine types are not specified. High-speed positioning within allowable positioning error ranges is required when using servo systems for positioning control. Vibration and other characteristics usually differ depending on the target position and travel distance, so it is necessary to establish optimal parameters to determine the desired speed and acceleration for each position and distance. However, it can be difficult to adjust large numbers of parameters.
Mitsubishi Electric previously developed products with 18 parameters in two categories, but has now developed a new system with 720 parameters in eight categories for highly precise control, where, it says, even experts would have difficulty adjusting such an extensive number of parameters.