Transforming Manual Processes with Computer Vision in Manufacturing - Part 1

Many manufacturing processes still rely heavily on manual labour for assembling and finishing components. However, with the integration of AI-powered computer vision, manufacturers can now gain actionable insights into these tasks, driving improvements in efficiency, consistency, and overall productivity.

Traditionally, measuring the efficiency of manual processes involves intermittent human observation and manual data collection. These methods are time-consuming, costly, and prone to inaccuracies, often yielding low-value data at the expense of non-productive work.

The solution? AI-driven systems and cameras that automatically monitor processes in real-time. With computer vision, manufacturers can delve into the fine details of their operations without needing continuous human oversight.

Computer vision is at the heart of this transformation. By capturing and analysing video streams of manual tasks, AI can track and assess various operations to pinpoint opportunities for optimising value-adding activities and minimising inefficiencies.

For example, consider a manual assembly process that has become a bottleneck in production. AI can identify non-productive time, such as when workers leave their stations to retrieve fasteners. By quantifying the impact of these delays, manufacturers can build a solid case for implementing solutions, such as improved workstation layouts. Similarly, if large variations in the time required for specific assembly tasks are detected, AI can help identify areas where design tweaks could streamline production and improve efficiency.

Computer vision’s potential goes beyond improving production line efficiency; it can be leveraged across the factory floor to streamline non-value-adding activities as well. A prime example is the monitoring of personal protective equipment (PPE) compliance, such as ensuring workers are wearing hard hats in designated areas. With AI-powered computer vision, manufacturers can automate the oversight of PPE usage in real-time, freeing up personnel and increasing compliance without manual intervention.

Cameras can be placed around the factory to monitor whether workers are wearing hard hats, automatically flagging any non-compliance. This ensures safety protocols are adhered to without the need for constant human observation, making it both cost-effective and efficient.

A key challenge remains however: how does the AI determine what's happening in the footage? This is where machine learning and data labelling come into play.

In order for an AI model to interpret video footage, the data must be labelled—meaning the system needs to know what it's looking at. For example, if the goal is to determine whether workers are wearing protective gear, such as hard hats, the AI needs to be trained on labelled images: some images where hard hats are present and others where they are absent.

Building a labelled dataset, however, can be a labour-intensive and tedious task. Typically, a human would have to go through frame-by-frame and label each image. For instance, one would label "yes" if a hard hat is present and "no" if it isn't. This process not only takes time but also costs resources, particularly when scaling up the operation.

A promising approach to overcome this hurdle is the use of existing AI tools, such as large language models, to automate the labelling process. By integrating these models into a system, manufacturers can prompt AI to analyse video footage in real-time. For example, you could feed video from your factory into the system and prompt the AI with questions like, "Is someone wearing a hard hat?" The AI would then return a simple "yes" or "no" based on what it sees in the footage. This allows manufacturers to generate labelled datasets without the need for human intervention.

Once the dataset is labelled, it can be used to train a machine learning model that can perform the task independently. In the case of monitoring hard hat usage, the system would be trained to recognise when workers are wearing hard hats based on the labelled images provided during the training phase. The end result is a machine learning model that can automatically analyse every frame of video and provide real-time insights into whether safety protocols are being followed.

This approach enables manufacturers to build custom AI models tailored to their specific operations. The initial use of a large language model is only needed to create the labelled dataset. After the model is trained, the process becomes self-sustaining, eliminating the need for ongoing AI involvement.

One of the standout features of this approach is that it doesn’t require massive, resource-intensive computing power. Once trained, these machine learning models can run on inexpensive hardware like a Raspberry Pi. This makes it an accessible and cost-effective solution for any size manufacturer.

A fully trained model can run continuously, providing real-time insights on manual processes, all while operating on low-power devices. This ensures that manufacturers can maintain operational efficiency without incurring high infrastructure costs.

This computer vision AI approach has broad applications in manufacturing, from monitoring safety compliance to improving workflows. By reducing the need for manual monitoring, manufacturers can focus on more strategic tasks while ensuring consistent quality and safety.

The beauty is that almost any manual process—whether it’s part inspection, assembly line work, or quality control—can benefit from automated AI-driven monitoring. By leveraging AI, manufacturers can gain deeper insights into their operations, enabling smarter, data-driven decision-making.

Computer vision AI offers manufacturers a powerful tool to automate the monitoring and analysis of manual processes. By using LLMs to label datasets and train machine learning models, manufacturers can create small, efficient systems that operate consistently and accurately. With the ability to deploy these models on low-cost hardware, the solution is both scalable and accessible to a wide range of operations. This approach exemplifies the future of manufacturing—where AI doesn’t just assist but becomes an integral part of optimising and enhancing industrial processes.

In the next part of this series, we'll walk through a real case study of computer vision AI in a manufacturing environment, presenting each stage of the user journey, from collecting data to training the model and, finally delivering actionable insights that transform operations. This case study will highlight the tangible benefits of AI and give you a clear roadmap for how it can be applied to your own processes.