Hardware Implementation of Convolutional Neural Network for High-Precision Machining at the Sensor Edge

The big data acquired from wireless sensor nodes for decision making on the cloud can overload the communication network adding to the total inference time. This paper presents an efficient hardware implementation of the artificial intelligence (AI) algorithms at the sensor edge for high-precision machining of mechanical parts, where the inference speed is critical. The integration of Edge AI shows synthesis and verification of convolutional neural network (CNN) on field-programmable gate array (FPGA) embedded in the sensor node significantly improves the decision-making time. The proposed 1D-CNN implementation on MAX10 FPGA series from Intel, at sensor edge can swiftly detect the cutting tool wear using the bending moment signals measured from embedded sensors on the tool holder. A miniaturized printed circuit board (PCB), hosting only the FPGA device to implement AI on the tool-holder is presented and compared with conventional wireless transmission of raw data for implementation of AI on cloud. The total decision-making time based on pre-trained CNN has been reduced from more than 200ms on cloud to 102ms on FPGA at sensor edge.