Industrial vision quality inspection on an automated production line consists of an imaging system, a light source system, an image acquisition and a processing system, motion control unit and display unit. Multiple industrial cameras, sensors and auxiliary light sources are installed on the production line. The sensors sense the products, the industrial cameras capture the image information of the products on the production line and transmit the image information to the image acquisition and processing system through the network, the edge hardware combined with the deep learning software processes, analyses and identifies the images and produces results, based on which other devices are controlled to reject unqualified products and other operations. At the same time, the industrial vision quality inspection system communicates with the cloud server, and together with other systems in the factory, the cloud platform data management is realized to create an intelligent factory.

The AI quality inspection system is a combination of a series of advanced software and hardware product technologies, requiring the deployment of trained AI models to edge computing modules, powerful algorithms combined with advanced image processing technology to achieve full control of production quality data. To achieve industrial vision edge computing and analysis, the Qiyang i.MX8M Plus core hardware is a good choice. The i.MX8M Plus processor features a high-performance CPU, energy-efficient GPU, strong arithmetic NPU and excellent ISP, which can be directly applied to build industrial vision edge devices.

Advantages

Edge machine learning

The QiYang IAC-IMX8MP-CM core board is based on the NXP i.MX8M Plus quad-core Cortex-A53 + single-core Cortex-M7 architecture processor. The processor is capable of implementing edge-side advanced machine learning reasoning in areas such as industry and IoT. The processor has a built-in NPU with an arithmetic power of up to 2.3TOPS and supports various AI frameworks such as tensorflow-lite, armNN, OpenCV and Caffe. With the high performance Cortex-A53 core combined with the NPU, edge devices will support machine learning and inference inputs with little or no human intervention to make decisions on the local end. With the i.MX8M Plus processor, machine learning workloads can be shifted to end devices at the edge of the network, reducing data transfer to and from the cloud, providing better latency and shorter response times while ensuring data reliability and security.

Image version processing

The processor has powerful multimedia capabilities with support for 1080p60, h.265 / 4, VP9, VP8 video decoding and 1080p60, h.265/4 video encoding, integrated dual ISP image signal processing and support for dual camera inputs for stereoscopic images. The processor's ISP is capable of complex image processing tasks such as fast multiple exposures with high dynamic range (HDR) and wide-angle fisheye distortion correction, as well as handling image enhancement to render clear images even in low-light conditions for accurate machine learning recognition and processing.

Industrial reliability

The i.MX8M Plus processor has excellent industrial features to meet the high quality and reliability requirements required for industrial applications. The processor supports ECC's DRAM controller, which effectively reduces the soft error rate (SER); integrated time-sensitive network TSN to guarantee data transmission at the right time; dual Gigabit Ethernet ports for stable network communication; integrated 800MHz Cortex-M7 co-processor to assist the main core, which can execute real-time processing systems and ensure low latency; two CAN-FD peripheral interfaces Connects to other systems and supports low latency communication; Supports industrial grade temperature width to adapt to complex industrial environments.

At present, Chinese manufacturing is being transform and upgrade from a manufacturer of quantity to a quality, the application fields of artificial intelligence and machine vision will be continued to broaden, industrial vision is being upgraded from 2D to 3D, from traditional industrial vision to AI industrial vision, from a signal product to a comprehensive solution.