Manufacturing has long relied on statistical process control (SPC) to optimise production processes, though these methods have inherent limitations. Sampling-based data collection paints an incomplete and potentially inaccurate picture of actual production processes.
It can also impede real-time feedback controls used to optimise operations in real time. Machine vision, in contrast, captures real-world data about physical products and processes, including detailed geometric measurements made with subpixel accuracy, surface quality assessments, positional data, color and contrast variations, and other characteristics.
Vision systems also reveal temporal trends across production runs. By pairing vision data with sophisticated analytics, manufacturers have already experienced demonstrable benefits, revealing correlations between process parameters and quality outcomes that statistical methods might otherwise miss.
The ability to capture, store, parse, and analyse vision data will only accelerate as it advances in pace with artificial intelligence (AI) (particularly deep learning models that are already used for more complex vision use cases), the Industrial Internet of Things (IIoT), and cloud-based technologies.
Such an approach enables intelligent operations, where processes continually improve over time. For example, a supplier to automotive OEMs was able to enhance the visual inspection of electric vehicle battery caps using deep learning machine vision. In another example, a vision provider for manufacturers is fusing 2D and 3D image processing for real-time analysis for its customers. The solution empowers manufacturers to inspect every component immediately after moulding—without slowing down production lines.
Volume, Integration, Processing
Today, several technical challenges have impeded vision-based analytics from delivering their full potential for improving throughput, yields, and quality in complex, changing, or distributed manufacturing processes. Key challenges among include:
- Data volumes: A single production line might generate terabytes of vision data daily, requiring significant storage and processing capabilities.
- Integration complexity: Integrating vision data with data from other real-time sources such as IIoT sensors, AI tools, and cloud platforms is difficult for legacy manufacturing execution systems (MES) and enterprise resource planning (ERP) platforms.
- Processing limitations: The real-time processing requirements of manufacturing operations often conflict with the computational demands of analysing massive datasets.
Though these challenges deserve consideration, none presents an insurmountable barrier to leveraging vision data more effectively for process optimisation. The following pages explore the factors that manufacturers must consider and address along this journey.
Key Considerations for Aggregating Data
Implementing vision analytics for process optimisation requires careful planning and execution to be successful. Engineering and quality teams should assess existing infrastructure such as network capacity, data storage capabilities, and processing resources to ensure they has the capacity to handle current and future vision analytics workloads.
Leveraging vision data for analytics requires multidisciplinary expertise bridging machine vision, data analysis, and manufacturing processes. Organisations may need to invest in training existing personnel or consult with specialised third-party experts. Numerous industry surveys remind us that manufacturers find it hard to hire skilled talent, and appropriate up and reskilling require appropriate investments of time and budget. However, newer machine vision solutions come with easier to use interfaces and ready-to-use AI models.
Finally, integrating data from diverse hardware and software platforms can be made simpler if sourced through a single supplier able to ensure seamless interoperability and compatibility across platforms while providing unified support and development roadmaps. In one real life example, a plastic packaging tracking provider selected machine vision software for its scalability and backwards compatibility with existing vision hardware in place. The result was an unprecedented 100% detection rate across six sites.
Capturing and Tailoring Machine Vision Data
Several types of machine vision data readily provide low-hanging fruit for supporting process optimisation. Systems that measure part dimensions, assess surface quality, or track temporal data all offer insights into upstream process variations, tool wear, and a host of other variables.
Three-dimensional imaging data introduces additional data sources for advanced analytics, such as form and fit analysis, assembly simulation, and flush and gap verification. For example, the automotive industry deploys 3D imaging to ensure that body panels align correctly to reduce both assembly time and aerodynamic drag. Nonvisible as well as multispectral and hyperspectral imaging methods multiply potential data sources further, revealing variations in material composition, stress patterns, or surface contamination.
Conclusion
Converting vision data into a source for analytics must balance the data’s immediate operational priorities with future analytical objectives to minimise disruption and ensure ROI. Different data architectures are geared for different businesses. Select one that is scaled—and scalable— accordingly to existing operational and analytical capabilities.
And selection of software analytics platforms also must accommodate current capabilities and grow with future requirements. If possible, leverage vision analytics in small pilot projects aimed at improving specific processes. For instance, a pilot project might focus on reducing defects in a particular assembly line by using vision data to adjust machine parameters in real time.
Organisational readiness extends beyond technological considerations. Successful implementation requires cross-functional collaboration between quality, production, IT, engineering and external partners.
Learn more here
