Cloud-Edge Collaborative Architecture Helps Build Smart Factory Vision AI Defect Detection Applications

Introduction

With the advent of the era of Industry 4.0, digital and intelligent transformation has become the only way for enterprises in the industrial field to maintain their core competitiveness. Industrial manufacturing involves many links and scenarios, and the ability to collect and process various production data is the key to determining its degree of automation. Building a reliable industrial IoT data access layer to provide real-time and reliable data sources for upper-layer platforms and applications for analysis and decision-making can greatly improve work efficiency.

In this special series of articles, we will combine EMQ's years of practical experience in serving customers in the industrial field, starting from the requirements of common scenarios in the industrial field such as energy consumption monitoring, predictive maintenance, and product quality traceability, and provide practitioners with targeted solutions. Program reference.

background

In tens of thousands of smart factories, the quality of a component is often related to whether the entire product is qualified, the qualification rate of the entire production line, the production efficiency and economic cost of the entire factory, and even the market competitiveness of the entire enterprise. Therefore, product quality inspection is a key link that production enterprise managers must pay attention to.

With the rapid development of industrial intelligence, visual AI defect detection technology has gradually matured and been widely used. The use of visual AI defect detection technology for industrial defect detection has the advantages of non-contact, high efficiency, low cost, and high degree of automation, and has inestimable value in detecting defects and preventing defective products.

Based on the visual AI defect detection technology, how to realize the "smartization" of zero-defect production and self-process optimization and upgrading in factories will challenge the existing technology in two dimensions. On the one hand, it is necessary to continuously train and optimize the AI algorithm model to improve the vision Detection technology coverage and accuracy; on the other hand, data can be automatically uploaded to the production execution system and enterprise cloud big data analysis platform, which is convenient for subsequent big data analysis to continuously optimize the process, improve the efficiency of the production line, and continuously improve the management mode , and finally realize the intelligent ability of the entire factory to self-correct and continuously improve.

Analysis of the status quo of visual AI defect detection technology

Lack of independent processing of events at the edge near the production line

Product defects are sometimes unavoidable in the process, sometimes caused by abnormal settings of production line equipment parameters, equipment failures or operation errors, etc. Once the visual AI defect detects a product defect, it is necessary to promptly notify the on-site engineer through sound and light and other alarm signals for troubleshooting. Or through the automated system of the production line to perform processes such as diversion and error correction to avoid causing greater economic losses. The triggering or execution process of the alarm signal is generally completed through the sound and light alarm or PLC execution. The edge computing capability needs to ensure the real-time, efficient and independent of the alarm event processing.

Difficulty in data isomerization and aggregation

In the production process of intelligent manufacturing, product quality data not only involves the image flow of defect detection, but also involves multi-source heterogeneous protocol data, production and operation-related business data and product design process data collection of on-site multi-process and multi-production line production equipment. And the docking of MES, WMS, ERP and other factory management systems. These devices or systems are located in different network environments, such as production networks, office networks or cloud platforms, etc. It is necessary to build an information channel to break through the barriers to data interaction between various devices and systems, and to conduct comprehensive perception and collection of relevant data. Visual AI defect detection and big data analysis of other production and business data.

New Trend: AI Algorithm Model Based on "Cloud-Edge" Architecture

The "cloud-edge" architecture has become a new trend in visual AI defect detection architecture. "Cloud" is set up in the factory-level information center or the headquarters of the group. It masters the functions of overall management and control. It can also select appropriate models for centralized training according to actual production needs, and then publish the trained models to "Edge" for nearby reasoning. And receive the inference results returned by it for storage and management; the "edge" is set up on each production line in the factory to perform front-end data collection, preprocessing and simple inference work, and under the control of the "cloud", the production line is Products undergo real-time defect detection.

EMQ Vision AI Defect Detection Solution

In response to the current situation of visual AI defect detection scenarios in the industrial field, EMQ provides a complete solution through cloud native technology and cloud-edge collaborative architecture to realize the connection of visual AI defect detection image streams and massive industrial equipment data in "production line-factory-group" , move, process, store and analyze.

The program mainly consists of the following products:

components	product name
Edge data acquisition software	Neuron - Industrial Protocol Gateway Software
Edge Broker	NanoMQ - Ultra Lightweight Edge MQTT Messaging Server
edge computing software	eKuiper - Ultra-Lightweight IoT Edge Data Streaming Analytics Engine
IoT access platform	EMQX BC - Cloud Native Distributed IoT Access Platform

1. EMQ provides visual AI defect detection data docking processing capabilities at the edge. eKuiper supports Rest, gRPC, and msgpack RPC services to connect visual AI defect detection data, obtain image streams of defective products, compress them in real time, and store them at the edge and aggregate them to the factory-level data center or cloud.
2. The edge end realizes the linkage of visual AI defect detection equipment and automation equipment. When the visual AI defect detection equipment detects product defects on the production line, it can directly issue commands to the sound and light alarm and PLC through Neuron to give alarm notification or perform triage and error correction. process.
3. Build a "production line-factory-group" image stream and a data transmission channel for massive industrial equipment. The data is aggregated through the EMQ edge computing platform, transmitted to the EMQX in the factory, and then bridged to the EMQX in the cloud, and then flows into the time series database and AI analysis application through its rule engine, which is a big data analysis based on the image data and business data of the whole group's factory defect detection. The application lays a foundation to realize the mining and application of data value such as production quality traceability and production process optimization.
4. Through this solution of EMQ, a complete and self-circulating cloud-edge integrated AI model training process can be built: the image streams at the edge are aggregated and persisted to the cloud in real time, and the cloud AI can conduct model training in a timely manner and periodically optimize the algorithm model and release it to the edge. At the same time, the inference results of the new model are aggregated and persisted in real time to prepare for further optimization and intelligence of the factory production process.

EMQ Architecture Advantages

Multi-dimensional data aggregation and logical processing capabilities

The EMQ overall solution can collect and reverse control the image stream and real-time data of PLC, non-standard automation equipment, various instrumentation, visual AI defect detection equipment in the factory, and can respond to data at all levels of edge terminal, factory MES system, and cloud center Logical operations, event stream processing requirements.

Multi-dimensional data persistence capabilities

Through the built-in rule engine function of ekuiper and EMQX, image streams and business data streams can be pushed to various databases in real time at the edge, factory-level information center and cloud, including InfluxDB, TimescaleDB, MySQL, PostgreSQL and other time series databases and Relational Database. EMQX supports database data writing performance of 100,000+TPS per second, which can meet the real-time storage of tens of millions of data points per second. EMQX integrates the data collected and analyzed by multi-terminal Neuron and eKuiper, pushes the data to the database and big data system for persistent storage, and builds a robust underlying data architecture for enterprises to build production quality analysis and optimization.

Cloud-side collaborative management improves enterprise IT level

EMQ's cloud-edge collaboration framework remotely manages many edge software such as Neuron and eKuiper. Regardless of whether the network between the cloud and the edge is in direct connection mode or penetration mode, parameter configuration, log viewing, real-time monitoring, etc. can be easily implemented. Remote management.

In addition, the solution uses KubeEdge to orchestrate and manage edge software, realizing functions such as high availability, remote deployment, software upgrade, and edge offline autonomy of edge software, realizing edge autonomy of applications, greatly improving the stability of the overall system, and reducing Operation and maintenance costs.

Epilogue

By building a data highway for image streams and business data streams to factory-level data centers and cloud centers, EMQ's solution for visual AI defect detection scenarios breaks the information silos between inspection systems and production line automation equipment, based on different business layers Provides corresponding logical analysis and data persistence capabilities for event processing requirements, provides a guarantee for enterprises to continuously optimize visual AI defect detection algorithms through AI model training, and continuously improve factory production processes and enterprise management models based on big data analysis, helping enterprises to achieve Digital transformation to enhance market competitiveness.

Copyright statement: This article is original by EMQ, please indicate the source when reprinting.

Original link: https://www.emqx.com/zh/blog/smart-factory-ai-defect-detection-solution