Articles
That verdict is confirmed by Rikki Coles, technical lead for edge computing at the Advanced Manufacturing Research Centre (AMRC) in Sheffield, after a 10-month collaboration with local engineering company Tinsley Bridge. The two have partnered in a project in which machining data from the company’s shop floor has been transmitted to the AMRC using cloud computing techniques and then analysed with the aid of AI-based software developed by the research organisation, which is part of the University of Sheffield.
Delving into data
According to Coles, the AMRC became involved after the manufacturer had carried out some initial work of its own to monitor power consumption by a CNC lathe making automotive parts.
That work had involved fitting a split-coil transformer to the cable feeding power to the cutting head to measure the voltage and hence indirectly the current it was drawing. But the company felt that the basic data had greater potential for yielding useful information than it could extract and so got in touch with the AMRC which has what Coles describes as “a strategy for showcasing AI in industrial settings”.
In response, a team from the AMRC added extra communications capability to the set-up, using the MQTT machine connectivity protocol to upload data to the research organisation’s IBM cloud computing platform, from where it could be accessed for further analysis. That further analysis was carried out using a custom AI program written in-house at the AMRC in the open-source Python language.
Coles says that analysis involved “time series classification” in which the power consumption data collected at the machine was compared with operator timesheets that recorded what parts were being made at any specific time to identify “what the signals for a particular part looked like”. That data was then fed into a “regression classifier algorithm” which was able to learn and remember the power consumption signature for each part – in other words the way power consumption varied during the different stages of the machining process. The signatures for different parts are, he says, “very distinct”.
As Coles explains, the value of that learned data is derived from the ability to compare it with new data fed into the AI system as a live feed from the machine in operation on the shop floor. That facilitates an accurate part count of manufactured items. But continuous monitoring in that way can also immediately signal when the process is deviating from the optimal level. Hence it has the potential to act as a realtime quality control and monitoring tool.
Now, Coles continues, this “supervised learning” technique in which the system is fed with pre-recorded reference data may be followed by further work involving “unsupervised learning” in which the system effectively teaches itself when exposed to the live data feed.
Sharing the benefits
So far the project has been confined to proving the feasibility of the techniques. Coles says that one point that has come out of the work is the surprisingly light computing resource it has required, with PC platforms able to provide all the necessary processing power. But the intention is to use the project as a demonstrator, possibly with a video screen displaying live data from the machine to show other firms the benefits of using AI as a support tool.
Content published by Professional Engineering does not necessarily represent the views of the Institution of Mechanical Engineers.