This IIoT-based use case has been deployed in a Chinese automotive O.E.M. as a support device to study environment effects on machine-tool precision at the tooltip. Specifically, this use case is oriented to crankshaft manufacturing in a machine similar to Figure 1.
It is well known that finite element analysis and other computer aided engineering (CAE) technologies are fundamental for machine design. However, none of these technologies is able to consider all the possible operating scenarios neither all the machine design details effects on the final product, just because there is a lack of computing power, time, or both. In this case, machine-tools designed to meet automotive powertrain manufacturing requirements at micron range need an extra help to contemplate those effects coming from the surroundings, i.e., environment, production, operators and others and improve its operation. This extra help could be useful to improve machine design or provide feedback to a control system to overcome those effects when are changing over time.
In this use case, it is important for the machine-tool end user to control the machine precision when the temperature changes. It is also interesting from the machine manufacturer point of view in order to identify possible design improvement over the product.
Therefore, a complete study of the temperature in the critical parts of the machine end its surroundings has to be made in order to understand from data-driven models, the effects. Once those are understood, the knowledge can be used as part of the control system that is able to compensate those changes and maintain the machine precision as required.
The use case is applied to a machine-tool that is part of a production line of crankshafts with a cycle time of 48 seconds.
To get the effects of temperature change over machine-tool performance in terms of precision during operation, the Aingura IIoT approach is to get different machine structure and fluids temperature, environment temperature and compared them to the tooltip position overtime, being the first step in terms of knowledge discovery to understand the relationship between variables. Then as a second step, the temperature readings are sent to the machine PLC the offset needed to maintain the machine precision.
In order to get the required data, a proprietary embedded edge device called Oberon, is connected to the machine, as shown in Figure 1, where variable values are acquired using different types of sensors fused together with machine numeric control values and then pre-processed and stored.
Although, temperature is a slow variable mainly because its natural inertia ruled by the materials, the control system has to react as fast as possible to compensate the tooltip deviation. In this case, the interesting part of the measurements is not the value but its delta, which changes overtime. This approach enables faster reaction times giving the actionable insight needed by the control. To improve this response, data stream multivariate approach is considered, where the temperature correlation could give better understanding and also allow some predictive capabilities on the system.
The data required for these type of analysis has the following details:
To provide a solution to this question, the machine learning approach, tailored by Aingura IIoT, used in this use case is divided in two phases: variable selection (feature subset selection), and multi-output regression. The variable selection phase is one of the most important parts of this analysis because the system should be able, as efficient as possible, to process variable data and provide the actionable insight needed by the machine to compensate itself.
Considering that all the variables are selected by experience, some of them could be redundant or even further, not all of them are relevant to the model. Therefore, this first stage needs to measure the variable level of significance having in mind the insight that has to be delivered, that is, which temperature measuring points is more related to the change in position of the tooltip. This feature selection is useful to reduce the amount of data that has to be processed, significantly reducing the response time and the need of complex communication infrastructure.
During the second step a relationship between continuous variables and multiple outputs has to be found, i.e., the temperature change related to the tooltip position. In this case, a Regression Random Forest algorithm has been applied, which is an ensemble algorithm that fits a decision trees over subsets averaging the result to improve the predictive accuracy.
As a result, Figure 5 shows highly related variables to the tooltip position, in this case, the environment temperature with a structural temperature located in the machine basement.
From this result, only few predictor variables from the original 9 has been selected, increasing the algorithm speed. From this point, the algorithm is able to predict tooltip deviation in order to compensate and give feedback to the control system.
The main benefit of these application is that better knowledge from the machine to the designers has been obtain and can impact the machine design in terms of materials used and their specification.
However, the most important benefit is to enable the compensation strategies of machine-tool behavior during production. This is important to increase crankshaft quality in terms of tolerance variation during thermal changes and machine availability. This self-adaptive capability is important to improve machine’s availability during minimum and maximum temperature operation (i.e. summer and winter), where the machine-tool has to stop until stable environmental temperature is reached. This stop could be up to 2 hours that equals to more than 80 crankshafts per day not produced. Additionally, depending on the manufacturing site, a stopped machine costs around $50k per hour.