As industrial praxis shows, in the case of complex monitoring systems there are frequent false alarms due to the dynamic and stochastic behaviour of the manufacturing process. Thus, such systems are not always accepted in operation; they are then shut off by the machine operator.
Many research centres around the world [BARS91, INAS93, MORI93, WECK95] have thus strove to make progress in the field of artificial intelligence useful for computer-integrated manufacturing in order to improve the availability, capacity, reliability and cost effectiveness of complex process monitoring systems with reproducible manufacturing quality.
By using neural networks, monitoring and diagnosis systems are to be constructed that perform a multi-dimensional parameter analysis with the help of measurements captured by multiple sensors.
Because of the properties of neural networks, such as high processing speeds and fault tolerance, as well as the ability to determine a relation between input and output variables independently by means of an iterative learning process, they are better adapted to the dynamic character of unstable manufacturing processes than conventional systems incapable of describing accidental process events.
Fig. 10-14 shows an example of a system structure of a neural network used for process monitoring in grinding. The parameters described above, such as the AE
effect
Possible neural network output variables could then be such information as “service life end reached”, “chattering detected”, “grinding burn detected” or “roughness too high”. A number of faults in the process can thereby be detected. This example should only provide a small glimpse into the possibilities of neural networks in process monitoring and should thus not be seen as exhaustive. Other systems, e. g. fuzzy logic or the combination of neural networks and fuzzy logic, can also provide alternatives in the development of intelligent monitoring systems.
[1] Geometric instability
From the machining of the component alone and independently of process parameters, roundness errors form on the component surface as a result of its engagement with the grinding wheel and its positioning at the control wheel and workrest support.
• Machine-dynamic instability
The grinding machine has different resonance ranges that can lead to regenerative chattering by means of process-contingent stimuli (for example, a choppy workpiece, imbalance, etc.). With the help of modal analyses common today, these frequency ranges can be detected and avoided in the actual process.
• Instability induced by the material removal process
Especially material removal rates that are too high can lead to undesirable vi-
[2] Real depth of cut depends on contact force, process parameters and tool topography.
• Tool follows workpiece surface.
• Low dimension and shape accuracy.