Neuromorphic computing for anomaly detection in a laser powder bed fusion process

TL;DR

This paper demonstrates the first use of spiking neural networks on neuromorphic hardware for detecting anomalies in laser powder bed fusion 3D printing, improving detection accuracy and enabling low-power edge computing.

Contribution

It introduces a novel application of SNNs on neuromorphic chips for real-time anomaly detection in additive manufacturing, with optimized spike latency for noise reduction.

Findings

Enhanced anomaly detection accuracy with spike latency adjustment

Successful implementation on neuromorphic Intel Loihi chip

Potential for low-power, edge-based manufacturing monitoring

Abstract

This study is the first application of spiking neural networks (SNNs) for anomaly detection in the Laser Powder Bed Fusion (LPBF) additive manufacturing process. The neural networks were used to identify print processing anomalies generated by dropping of laser energy during the printing of individual layers in a Ti-6Al-4V alloy lattice structures. Associated changes in the laser generated melt pool were observed using an in-process photodiode monitoring technique. photodiode sensors capturing plasma and infrared radiations reflected from the print bed of the metal 3D printer were utilized to detect sudden changes caused by anomalies during the printing process. The algorithm is first implemented on non-neuromorphic hardware including a central processing unit (CPU), on Field Programmable Gate Arrays (FPGA) and then on neuromorphic Intel's Loihi chip. Improved detection of anomalies is achieved by adjusting the spike latency of the neural network, which reduces masking of information by noise within the monitored temporal signal. The work demonstrates the possibility of using low-power neuromorphic chips within an edge framework for anomaly detection in additive manufacturing and creates a framework for the process.