2. TOWARDS AN INLINE QUALITY MONITORING FOR
CRIMPING PROCESSES UTILIZING MACHINE LEARNING
TECHNIQUES.
M O R I T Z M E I N E R S , A N D R E A S M AY R , M A R L E N E K U H N , B E R N H A R D R A A B , J Ö R G F R A N K E I N S T I T U T E O F
FA C T O RY A U T O M AT I O N A N D P R O D U C T I O N S Y S T E M S ( FA P S )
3. Problem Identification
Crimp curve monitoring is a traditional approach used by manufacturers to assess the quality of
crimp connections. It involves comparing the actual curve progression of a crimp connection with
predefined reference values to determine its acceptability. However, this method offers only a
basic evaluation and lacks the ability to provide detailed insights into specific error patterns or
underlying conditions that might be contributing to issues.
In contrast, leveraging machine learning techniques can enable a more comprehensive analysis of
the crimping process, leading to the identification of precise issues and facilitating improvements
in the overall quality of crimp connections.
4. Methodology
The experimental setup involved using a Schleuniger Crimp Center 36SP with an F-crimp for the
experiment. The defined acceptable process specifications and a double-ended crimped wire
produced under those specifications were provided. The dataset consisted of six simulated
conditions, and the number of crimping operations performed for each condition was documented.
Crimp force curves were measured and analyzed.
To identify acceptable processes and detect anomalies, unsupervised learning techniques such as
Autoencoder (AE) and Variational Autoencoder (VAE) were employed. These models were trained
on the data from acceptable processes and learned to reconstruct similar data. Various performance
metrics were evaluated to assess the effectiveness of these models.
For process diagnosis and classification, a supervised learning approach using a 1D-CNN model
was utilized. The architecture and hyperparameters of the model were specified in detail. Multiple
models were trained and assessed using metrics such as accuracy, recall, precision, and confusion
matrices.
5. Result and conclusion
In terms of unsupervised learning, the VAE outperformed the AE, demonstrating superior
performance and less variation. Specifically, the VAE exhibited higher accuracy in detecting
anomalies within the less complex dataset. On the other hand, the AE excelled in categorizing
uncritical conditions for press 1.
Regarding supervised learning, the 1D-CNN model applied to press 1 achieved commendable
recall and accuracy in identifying critical anomalies. However, it faced challenges in reliably
identifying temperate wire and loose wire inlet conditions. The 1D-CNN model for press 2
displayed similar capabilities in detecting anomalies overall.
The utilization of machine learning models enables comprehensive monitoring of the crimping
process by classifying crimp connections as either acceptable or unacceptable. This advancement
facilitates error correction and traceability in wire harness assembly, which plays a crucial role in
the development of autonomous and electrified vehicles.
6. APPLICATION OF THE LEAN CONCEPT FOR ANALYSIS AND
OPTIMIZATION OF THE AUTOMOBILE FILTER PRODUCTION
COMPANY
S O F T I C A , B A S I C H , L U L I C H , N A K I C A .
7. Problem Identification
Automobile filter production companies encounter challenges related to inefficient production
processes. These challenges encompass various aspects such as excessive wait times, inefficient
handling or transportation of materials, unnecessary inventory, and bottlenecks in the production
flow. The presence of these inefficiencies results in longer lead times, decreased productivity, and
elevated costs.
8. Methodology
The objectives of the analysis and optimization process encompass various goals, such as
enhancing productivity, reducing lead times, minimizing defects, and improving customer
satisfaction. These objectives may also include specific targets that are relevant to the organization.
The implementation of methodologies such as 5S (Sort, Set in Order, Shine, Standardize, Sustain),
standardized work, Kanban systems, single-piece flow, mistake-proofing (Poka-Yoke), Total
Productive Maintenance (TPM), and other strategies can be employed to achieve these objectives.
9. Result and conclusion
Through the implementation of Lean principles, the company successfully eliminated
transportation waste by enhancing machine layouts and optimizing workplace ergonomics. The
identification and resolution of bottlenecks and process halts were achieved by procuring new
equipment and modifying the sequence of operations. Furthermore, the company established a
standardized sequence of operation and implemented standardized work tasks to ensure consistent
and precise results from the collected data, enabling effective monitoring and analysis.