Smart sheet metal forming: importance of data acquisition, preprocessing and transformation on the performance of a multiclass support vector machine for predicting wear states during blanking

Abstract

In consequence of high cost pressure and the progressive globalization of markets, blanking, which represents the most economical process in the value chain of manufacturing companies, is particularly dependent on reducing machine downtimes and increasing the degree of utilization. For this purpose, it is necessary to be able to make a real-time prediction about the current and future process conditions even at high production rates. Therefore, this study investigates the influence of data acquisition, preprocessing and transformation on the performance of a multiclass support vector machine to classify abrasive wear states during blanking based on force signals. The performance of the model was quantitatively evaluated based on the model accuracy and the separability of the classes. As a result, it was shown, that the deviation of time series represents the key parameter for the resulting performance of the classification model and strongly depends on the sensor type and position, the preprocessing procedure as well as the feature extraction and selection. Furthermore, it is shown that the consideration of domain knowledge in the phases of data acquisition, preprocessing and transformation improves the performance of the classification model and is essential to successfully implement AI projects. Summarizing the findings of this study, trustworthy data sets play a crucial role for implementing an automated process monitoring as a basis for resilient manufacturing systems.

Perspectives on data-driven models and its potentials in metal forming and blanking technologies

“…Here, explainability refers to the amount of reliable gain of knowledge and allows deeper insights into the phenomena acting in processes [7]. While common black-box models like regressions and support vector machines (SVM) are comprehensible and therefore widely used in the production engineering environment, the results generated by a neural network training are usually no longer transparent for the Blanking (force and displacement) [41] Deep drawing (AE) [44] Blanking (acceleration and force) [47] Forging (force) [50] Stamping (force) [51] Roll forming (temperature and force) [42] Blanking (acceleration and force) [45] Progressive stamping (force) [48] Fine blanking (force) [52] Blanking (force and AE) [43] Deep drawing (AE) [46] Blanking (acceleration and force) [49] Blanking (force) [53] Feature selection…”

“…However, predicting the amount of wear at cutting tools during series production is currently not possible. In this context and in a preliminary work, a black-box model based on a support vector machine (SVM), which allows for prediction of current wear state based on ´in-process force-signals´ was developed [53]. Therefore, five abrasive wear states were characterised by the increase of cutting edge radii r i of the blanking tool during processing and are estimated by a classification model based on a SVM.…”

Relating wear stages in sheet metal forming based on short- and long-term force signal variations

“…In addition, they have been shown to contain information about the progression of wear (Voss et al, 2017), quality characteristics of the resulting workpieces (Havinga & Van Den Boogaard, 2017), machine errors, feed errors, and thickness variations (Bassiuny et al, 2007). The high sensitivity of force signals to changes in punch wear condition has also recently been demonstrated by the use of classification techniques to correctly evaluate the wear condition of the tool (Kubik et al, 2022) at a given time, while also a continuous monitoring approach has been introduced through the use of autoencoder (AE) (Niemietz et al, 2021). In most studies, the extraction of meaningful features from high-dimensional time series data has been shown to be a critical step in aggregating information about the physical state of the process.…”

“…The summary of the research results suggests that force signals can be used to provide adequate insight into the process and to develop applications for monitoring and quality prediction. However, while it was shown in (Kubik et al, 2022) that force data can be attributed to various punch wear states, no model could be found that links the continuous wear evolution to quantifiable indicators such as trends, variations or events in force data over large stroke series. This is of particular importance, as a monitoring system should not only be able to distinguish different wear states, but also to relate different wear states to each other in a certain metric.…”

Towards a real-time tool state detection in sheet metal forming processes validated by wear classification during blanking

“…However, manually processing of the huge amount of available data in high speed blanking processes complicates this knowledge discovery procedure and requires advanced analysis techniques such as machine learning (ML) [5] or featurebased supervision (FS) [6]. In the context of in-line wear estimation during blanking, DBM approaches such as ML or FS are able to quantify complex and non-linear interdependencies between wear phenomena and a sensorial acquired variable [7].…”