Process monitoring systems offer the opportunity for in process decisions to be shifted from human operators to automated systems to facilitate multi-machine manning
The high cost of using skilled operators in production processes has created a demand for reduced manning manufacture. Hence process monitoring has been widely researched as there is a clear need for intelligent systems to replace manual intervention in existing processes.
Tool condition monitoring is often the main reason for monitoring a machining process. Although, several other research areas have considered a fault detection approach where detection of chatter, depth of cut change or cutting force changes are the objective of the system.
That said, detailed understanding of the relationship between input variables and the measured effects are still lacking. Current methods require further experimental data and systems training in order to be effective on unproven processes.
The majority of processmonitoring applications target the detection of process failures rather than the detection of the cause of failure. With the functionality to detect cause of failures, a monitoring system is more capable of preventing issues in the manufacturing process since a cause of failure must be known to effectively correct the process. In addition to this, identifying the effects that may occur from a failure is also beneficial – the most commonly used method being on-machine probing for part geometry error measurement.
Failure Mode Effects Analysis (FMEA) was used to examine existing machining processes to determine the failure that occurred, cause of the failures, effects from the occurrence and the detection methods currently in use.
This identified that the operator was responsible for detecting many failures. For the responsibility to be shifted from the operator, a monitoring system needed to be built that would be able to detect each one of the failures.
A common failure shown in the FMEA is a prematurely worn tool. A root cause of this failure may be material property change and an effect of this failure may be poor surface finish. However, several other interactions occur during this process that result from the root cause (a metacause) such as increased cutting force.
Meta-causes generally occur during the machining process from which indirect measurement can be taken (i.e. vibration, acoustic emission and spindle power measurement).
The model was used to create a system with the ability to measure meta-causes and diagnose root causes. The defect(s) would then be anticipated and either measured in process (where a method to do so exists) or the user would be informed of the potential defect. The system would be retrospectively informed of the defect, either from measurement equipment or an operator input. Then it would then infer the most likely root cause of this error from the previous meta-cause measurement data.
Most significantly, the system will be self-learning and therefore, it can improve its diagnosis performance based on previous results.
A process monitoring system was defined and tested during a profile milling operation on titanium 6-4. From the results obtained, a graph was produced howing the predicted depth of cut and tool condition having a good fit to the actual values.
Using both spindle power and acoustic emission signal magnitude, the model obtained from experimental data can clearly differentiate between the change in depth of cut and change in tool condition using these sensor signals alone. Furthermore, depth of cut fits to the actual values accurately, within 0.1mm using this system. Tool life has been assessed by the amount of work done by the tool (metal removed) and gives a fit to within 10-20% of the tools life.
Overall, the methodology described enables fault detection systems to be designed and built with minimal expense. Customised systems can therefore be implemented and tested on many production scenarios, providing a more flexible tool for process monitoring than is used in current aerospace manufacturing processes.