Robin Ströbel

The rapid individualization of milling operations have introduced unprecedented complexity and variant diversity, necessitating adaptive process monitoring under scarce-data conditions. This systematic literature review (SLR) takes a comprehensive look at machine learning (ML)-based approaches for process monitoring and anomaly detection in highly flexible milling processes, focusing on single-part and rapidly changeing production scenarios. The fourteen most relevant studies published since 2019 were identified by adhering to established SLR frameworks. The methods are evaluated in terms of their flexibility, data efficiency, model quality and cost-effectiveness. It is revealed by the SLR that transfer learning (TL), physics-informed ML and active learning (AL) are frequently used to address the issue of limited training data whilst improving the robustness of the model. However, there a shortcoming in the integration of multiple data-efficient training strategies within holistic frameworks. Additionally, focusing on internal machine signals could reduce the burden of monitoring systems on brownfield machines. Combining monitoring via internal machine signals with AL, TL, physics-informed ML and data augmentation offers promising research directions for scalable, low-cost process monitoring in flexible manufacturing environments.

Machine acceptance is a vital part of the manufacturing process, especially for 5-axis machine tools prevalent in the aerospace industry. It is currently done by skilled workers using their experience and knowledge to iteratively improve the machine tool until it is able to manufacture a test piece that meets the required quality standards. This process is time consuming, requires a lot of expertise, and is not easily transferable to new workers. In this paper, we propose a system that uses machine control signals to detect anomalies during the manufacturing of the test piece and classify them by their cause, like an onset of chatter, positional errors, or others. For this, the machine signals are segmented using a sliding window approach. Multiple strategies to reduce the dimensionality of the segments are evaluated, including autoencoders based on a Convolutional Neural Network or a Long-Short Term Memory Network as well as manually designed features. The reduced segments are then classified using a Random Forrest. The results show that the proposed system is able to detect anomalies with high accuracy and classify them correctly.

The transition to highly customized, one-off production in modern manufacturing necessitates sophisticated process monitoring to reduce waste, minimise downtime, and alleviate operator burden. Computer Numerically Controlled (CNC) axes represent a fundamental component of automated manufacturing and offer a universal and accessible monitoring option through power supply data. By accurately predicting reference signals and comparing them with real-time measurements, deviations can be used for effective model-based process monitoring and anomaly detection. This study explores the efficacy of hybrid machine learning (ML) models in predicting reference signals for CNC axes using features derived from a physical model. Additionally, relevant but difficult-to-measure features such as process forces and the material removal rate (MMR) were made accessible through soft sensors. Various ML models were evaluated, including tree-based models (e.g. random forest (RF) and gradient boosting (GB)) and deep learning (DL) models (e.g. feed-forward neural networks (FNN), long short-term memory (LSTM), and transformers-based models (TF)). Feature importance analysis was performed, identifying velocity, acceleration, process forces, spindle torque, andMMRas crucial predictors that influence model performance. Key results indicate that tree-based models, specifically RF and GB, consistently delivered the highest accuracy, achieving R2 up to 0.98 for translatory axes and approximately 0.89 for the main spindle. These models demonstrated robustness, outperforming deep learning approaches, particularly when trained on smaller datasets. Although DL models improved with larger data volumes, their performance remained inferior compared to tree-based methods. The study underscores the potential of the integration of physical knowledge into hybrid ML models to enhance model-based process monitoring.

In einer individualisierten Produktion mit hoher Variantenvielfalt stoßen traditionelle Ansätze zur Prozessüberwachung immer häufiger an ihre Grenzen. Diese basieren meist auf statischen Datensätzen oder wiederkehrenden Prozessmustern, was in dynamischen Produktionsumgebungen zu ungenauen Vorhersagen und vermehrten Fehlalarmen führen kann. In diesem Beitrag wird ein Konzept zur Flexibilisierung der Prozessüberwachung diskreter Produktionsprozesse vorgestellt, das auf der Kombination von Physics-informed Machine Learning (PIML) und Active Learning (AL) basiert. In agilen Produktionsumgebungen können so nicht nur Anomalien erkannt, sondern das Überwachungsmodell bei Fehlalarmen auch automatisch aktualisiert werden. Dadurch bleibt das Überwachungssystem auch unter variablen Produktionsbedingungen präzise, was Fehlalarme reduziert und damit zu einer verbesserten Overall Equipment Effectiveness (OEE) beiträgt.

This study addresses a significant gap in the field of time series regression modeling by highlighting the central role of data augmentation in improving model accuracy. The primary objective is to present a detailed methodology for systematic sampling of training datasets through data augmentation to improve the accuracy of time series regression models. Therefore, different augmentation techniques are compared to evaluate their impact on model accuracy across different datasets and model architectures. In addition, this research highlights the need for a standardized approach to creating training datasets using multiple augmentation methods. The lack of a clear framework hinders the easy integration of data augmentation into time series regression pipelines. Our systematic methodology promotes model accuracy while providing a robust foundation for practitioners to seamlessly integrate data augmentation into their modeling practices. The effectiveness of our approach is demonstrated using process data from two milling machines. Experiments show that the optimized training dataset improves the generalization ability of machine learning models in 86.67% of the evaluated scenarios. However, the prediction accuracy of models trained on a sufficient dataset remains largely unaffected. Based on these results, sophisticated sampling strategies such as Quadratic Weighting of multiple augmentation approaches may be beneficial.

This study addresses the challenge of the optimization of milling in industrial production, focusing on developing and applying a novel framework for optimising manufacturing processes. Recognising a gap in current methods, the research primarily targets the underutilisation of advanced data analysis and machine learning techniques in industrial settings. The proposed framework integrates these technologies to refine machining parameters more effectively than conventional approaches. The research method involved the development of the framework for the realisation and analysis of measurement data from milling machines, focusing on six machine parts and employing a machine learning system for optimization and evaluation. The developed and realised framework in the form of a software demonstrator showed its applicability in different experiments. This research enables easy deployment of data-driven techniques for sustainable industrial practices, highlighting the potential of this framework for transforming manufacturing processes.

In today’s manufacturing landscape, Digital Twins play a pivotal role in optimising processes and deriving actionable insights that extend beyond on-site calculations. These dynamic representations of systems demand real-time data on the actual state of machinery, rather than static images depicting idealized configurations. This paper presents a novel approach for monitoring tool and component wear in CNC milling machines by segmenting and classifying individual machining cycles. The method assumes recurring sequences, even with a batch size of 1, and considers a progressive increase in tool wear between cycles. The algorithms effectively segment and classify cycles based on path length, spindle speed and cycle duration. The tool condition index for each cycle is determined by considering all axis signals, with upper and lower thresholds established for quantifying tool conditions. The same approach is adapted to predict component wear progression in machine tools, ensuring robust condition determination. A percentage-based component state description is achieved by comparing it to the corresponding Tool Condition Codes (TCC) range. This method provides a four-class estimation of the component state. The approach has demonstrated robustness in various validation cases.