How are Quantum Techniques Improving Additive Manufacturing?

By Dr Silpaja Chandrasekar, PhDReviewed by Susha Cheriyedath, M.Sc.Nov 5 2024

In a recent Manufacturing Letters paper, researchers present a pioneering quantum-enhanced approach to address machine learning constraints in advanced manufacturing.

By deploying quantum support vector machines, the team effectively identified machine states in fused filament fabrication through acoustic data analysis, marking a significant step forward in predictive maintenance for additive processes.

Additionally, the study applied quantum convolutional neural networks (CNNs) to pinpoint spatter events in laser powder bed fusion using optical imagery, demonstrating that quantum models can achieve classical levels of accuracy with exponentially reduced parameter requirements, optimizing computational efficiency for complex manufacturing tasks.

Quantum ML for Advanced Manufacturing

In this study, a quantum support vector machine (QSVM) was used to enhance anomaly detection within the fused filament fabrication (FFF) process. By leveraging acoustic emission (AE) sensor data, QSVM was able to effectively distinguish between critical machine states such as 'normal,' 'run-out-of-material,' 'semi-blocked,' and 'blocked.' The classification was based on analyzing AE data attributes, including acoustic burst signal (ABS)-energy, root mean square (RMS), and AE counts, which are fundamental indicators of operational anomalies in additive processes.

A 500 millisecond time window facilitated the calculation of the mean and standard deviation for each attribute, yielding six distinct features. QSVM was then trained using 75 % of the dataset, while the remaining 25 % was reserved for testing. The results revealed discernible patterns, especially for the 'blocked' and 'run-out-of-material' states, underscoring QSVM’s capability to enhance real-time process monitoring.

The QSVM implementation utilized a quantum feature map based on the Havlíček et al. approach, encoded with the ZZ feature map using Qiskit. Encoding was optimized with four qubits for the RMS and ABS-Energy attributes in the initial model, and expanded to six qubits to include AE counts in the extended configuration.

QSVM’s performance, benchmarked against classical SVMs with linear and RBF kernels, demonstrated competitive accuracy while capitalizing on the advantages of quantum feature mapping.

The study also introduced a quantum convolutional neural network (QCNN) tailored for spatter detection in laser powder bed fusion (LPBF). High-speed imaging captured the LPBF melt pool at 1000 frames per second, and frames were then scrutinized and labeled as either 'Spatter' or 'No Spatter.' The QCNN model processed cropped and resized melt pool images, encoding them with 8 qubits for focused analysis of spatter phenomena.

The QCNN architecture incorporated quantum convolution and pooling layers, which optimized data retention while minimizing qubit requirements. Rotation and controlled-NOT (CNOT) gates within the convolution layers enhanced feature extraction, while a quantum pooling mechanism reduced computational demand by condensing two qubits into one.

Training was performed over 100 epochs with the COBYLA optimizer, and the QCNN’s performance was assessed against a classical neural network of comparable architecture. QCNN’s efficiency in parameter reduction and competitive accuracy underscored the transformative potential of quantum machine learning for advanced manufacturing, paving the way for more efficient, data-driven insights in high-stakes manufacturing environments.

Quantum Anomaly Detection

This study explored the application of quantum machine learning (ML) techniques, particularly the QSVM and QCNN, to enhance anomaly detection in FFF and LPBF processes.

The QSVM model demonstrated a strong performance, achieving 85.7 % accuracy with four features and 89.5 % with six features, thereby outperforming classical SVMs with linear and RBF kernels. This improvement in classification accuracy with additional features underscores the efficacy of mapping data into high-dimensional Hilbert space, a distinct advantage of quantum approaches in capturing complex data patterns.

In parallel, the QCNN effectively detected spatter events in LPBF, achieving a training accuracy of 75.0 % and a test accuracy of 64.6 % using only 63 parameters, achieving comparable accuracy to classical neural networks but with exponentially reduced parameter requirements. The QCNN’s architecture, designed as an n-level binary tree, provided a marked improvement in efficiency, reducing space complexity significantly compared to traditional neural networks.

The findings suggest that both QSVM and QCNN architectures offer resilience against overfitting, a critical concern in manufacturing anomaly detection where training datasets are often limited. Additionally, the study highlighted the unique sampling complexity intrinsic to quantum algorithms, a factor that complements the conventional time and space considerations of classical ML approaches.

Leveraging quantum-specific feature mapping and parameter efficiency, these quantum ML methods present a promising avenue for advanced manufacturing applications, enabling more precise and resource-efficient anomaly detection and classification.

Conclusion

In summary, this study showcased the potential of QML to enhance manufacturing applications, demonstrating QSVM’s effectiveness in identifying critical machine states in FFF and QCNN’s capability in spatter detection within LPBF processes.

The results highlighted QSVM’s quantum kernel for superior mapping to high-dimensional feature spaces, while QCNN achieved accuracy comparable to classical neural networks with a fraction of the parameters, effectively reducing overfitting risks.

Although current quantum computing technology poses limitations, the study suggests that custom-tailored quantum ML models hold promise for addressing scalability challenges and advancing predictive accuracy in dynamic manufacturing environments.

Journal Reference

Choi, E., et al. (2024). Quantum machine learning for additive manufacturing process monitoring. Manufacturing Letters, 41, 1415-1422. DOI: 10.1016/j.mfglet.2024.09.168, https://www.sciencedirect.com/science/article/pii/S221384632400258X

Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.