This is part three of a blog series on video recognition using deep learning algorithms. The previous articles in this series were Video Recognition: Real-Time Object Detection and Classification and Video Recognition using Deep Learning – Deep Dive.
In this article, we will explore how these algorithms can enhance safety and productivity in industrial settings using factory floor video feeds.
What areas of the value chain are impacted?
Let us first consider worker safety. Industrial workers are almost always exposed to hazardous working conditions such as high temperature, pressure, and airflow. Estimating and minimizing occupational risk attributed to these conditions depends greatly on how the manufacturing and material handling workforce activity is in areas where such conditions are prevalent. Accordingly, as per statutory guidelines, proximity alert lines are marked around assembly line machinery as countermeasures to ensure safe industrial operations.
Video surveillance systems with cameras purposively directed at workstations throughout the factory feeding into ML-based video recognition solutions help monitor events like intentional and inadvertent triggering of these countermeasures and subsequently alert the workers, supervisors, and operations managers. They also help create heat maps that can profile this aggregate activity so that assembly line jobs and material routes can be designed optimally to improve throughput, reduce risk, and finally improve occupational safety standards compliance.
The next perspective is manufacturing process excellence for continuous improvement.
One approach is to automate the entire manufacturing process in the factory to create dark factories using lights-out manufacturing technology discipline. However, there are still possible situations of transmission failures e.g., conveyor belt deviation, slippage, and blockage, or even belt breakage, that leads to wastage. ML-based video analysis systems can detect these failures in highly occluded, complex machine assemblies on minimally illuminated, practically dark factory floors.
DOWNLOAD CASE STUDY
Face on Edge-Realtime Face Recognition
If lights-out manufacturing is not viable, another way is to improve worker efficiency by improving the manufacturing job routines and standard process adherence. If an assembly line operator is improperly operating a machine tool, not ensuring sufficient inventory levels, or using incorrect machine configurations, it can affect the manufacturing throughput. Video analytics helps analyze operator routines, detect common bottlenecks and inefficiencies, and analyze the root cause for these. Assembly lines can be accordingly optimized, and throughput enhanced.
Quality check on finished products is another key activity. If fully automated and optimally designed, it is a deterministic process. When human operators are involved, there is a stochastic element to it, attributed to individual ergonomics, strength, and fatigue that may come from working repetitive tasks. This makes monitoring the QC station operator activities, with reference to standard operating procedures and checklists, an important consideration.
Human action recognition helps us in two ways. First, in breaking down the manual QC process in the video feeds into atomic actions; and second, analysis of these actions to ensure adherence to a set of predefined sequence of actions as per the training data. Subsequently, operator score over a period can be calculated and training assignment or SOP correction can be initiated. Recorded video feeds can be used as training aids for diverse roles across the manufacturing value chain.
Industrial asset security and access management is vital to ensuring manufacturing operation integrity and control. Various workforce roles require access to these assets at varying times in the manufacturing process, for different periods, and at different frequencies. Video recognition systems with face recognition, object detection, classification, and instance counting help in ensuring that only authorized personnel have access to industrial assets. Further, the extent of access can be scrutinized and violations of access guidelines are captured, monitored, and analyzed for corrective actions.
Along with the manufacturing assembly line workforce, material management workers and supply chain workers are critical components of the manufacturing value chain, moving material at various stages of manufacturing from inventory to staging areas on floors to individual workstations and back. Safe and streamlined navigation of material handling vehicles, manned, unmanned, or autonomous, in challenging industrial environments are a key success factor for optimized throughput of the manufacturing value chain. Key use cases include driver assistance for collision awareness and emotion detection for signs of fatigue, heightened stress.
Deep computer vision algorithms on live video feeds are often deployed in these situations, leading to reduced risk for the drivers, material in motion, and the factory workforce at large.
Video recognition and operational analytics is only the first step in a broader Industry 4.0 roadmap. Adopting technologies like digital twins, extended reality, and intelligent process automation to complement AI-powered machine vision promises to unlock order of magnitude improvements in manufacturing process efficiency and effectiveness.
eInfochips offers comprehensive computer vision solutions for diverse industry verticals like transportation, industrial, pharmaceutical, smart cities and consumer electronics across the model lifecycle from algorithm selection, training, validation, inferencing, deployment and sustenance. For more information please contact us today.