USING MACHINE VISION ANALYTICS TO INFORM PROCESS OPTIMIZATION

Machine vision systems are often utilized to inspect products. When connected to machine vision cameras on a factory floor, software can detect visual anomalies such as cracks in a pipe, or tablets or medicine that are the wrong color. Some software can also monitor processes and detect patterns that lead to errors, inefficiencies, and defects. When a manufacturer can identify the root cause of an issue, they can work to correct it by adjusting the process.

In a machine vision solution, a computer receives visual input from cameras and software analyzes what it is seeing. Then, using rules-based algorithms or AI-powered software, the system can act on the information — with or without human intervention. In manufacturing, machine vision has many potential applications, from quality control inspections to safety and security monitoring.

Fig 1: Machine vision systems can monitor procedures in areas such as printed circuit board assembly to help optimize the process.

Machine vision can also be used to optimize processes and maximize use of resources. For example, cameras capture images of packaged goods, such as bottles of lotion in a consumer goods manufacturing plant, and send them to software that recognizes irregularities like broken safety seals or missing labels.

With machine vision software, machines can analyze and make decisions without the need for constant human oversight. But such solutions only perform well if the models are trained using high-quality data, which means producing high-quality images. “Garbage in, garbage out” is an apt phrase here. Whether the system is based on discrete, rules-based algorithms or machine learning, high-quality images will help ensure success. Emergent’s line of 10GigE, 25GigE, and 100GigE machine vision cameras deliver zero-data-loss imaging capabilities and can be deployed with GPUDirect technology, which enables the transfer of images directly to GPU memory. GPUDirect offers zero CPU utilization and zero memory bandwidth imaging, and can be leveraged through Emergent’s eCapture Pro software.

Fig 2: Adding machine vision cameras to robots adds significant flexibility, allowing the system to not only monitor a process such as solar panel assembly, but also to guide the robot for precise movements.

As a machine vision solution monitors a process, such as a factory assembly line, it collects large amounts of data that can be used to create a baseline for what normally occurs during that process. The software applies intelligence to increase efficiencies and reduce anomalies. A few examples include:

• Removing faulty raw materials Machine vision software can recognize visual defects in products early in a manufacturing process. For example, in food production, the machine vision solution might utilize near infrared (NIR) or shortwave infrared (SWIR) cameras, such as Emergent’s area-scan cameras, to recognize high water density in spoiled produce.
• Recognizing problematic processes Manufacturers can use machine vision to observe and evaluate patterns in existing processes, such as applying paint to the body of a car in an automotive manufacturing plant. As cameras capture the processes of cleaning, priming, sealing, and painting, the software compiles and analyzes visual data to trace root causes of machine failures and defects. In conjunction with other connected sensors, machine vision can also be utilized for predictive maintenance — for example, recognizing when a piece of equipment is vibrating too much — helping manufacturers fix issues before they cause downtime.
• Continuous process improvement Over time, a machine vision solution’s algorithm can be trained to become better at spotting anomalies and inefficiencies. Machine learning software can learn the nuances of a manufacturer’s particular process and become better equipped to make targeted adjustments and suggestions for optimization.

A wide range of machine vision solutions on today’s market can be utilized for process optimization. Depending on the use case, a vision solution will need to examine different forms of analog and digital signals. Most solutions fall into one of the following categories:

2D machine vision (area scan imaging)
The current default for most machine vision applications, two-dimensional machine vision solutions take standard 2D images of objects, which work very well for inspecting single-part items and scanning barcodes. Area scan cameras feature rectangular image sensors that capture images in one frame, with the resulting image corresponding to the height and width of the image sensor. Emergent offers an entire ecosystem of high-speed area scan cameras in its HR 10GigE, Bolt 25GigE, and Zenith 100GigE cameras. Sensor options include the latest from Gpixel and Sony’s Pregius S series and range from 0.5MP to 100MP+ with speeds up to 3462fps.

2D machine vision (line scan imaging)
Line scan cameras, meanwhile, use a linear image sensor to build an image one pixel row at a time at high speeds. In a line scan vision system, the object being inspected moves on a conveyor or similar device under the camera to capture the whole surface one pixel row at a time. Emergent offers several line scan camera models in 10GigE, 25GigE, and 100GigE formats based on Gpixel CMOS line scan sensors ranging from the 4Kx2 GL3504 to the 16Kx16 GL5016. Line rates range from 70KHz single line rate in the 10GigE Pace LR-16KG35 up to 400KHz in the 100GigE Pinnacle LZ-16KG5.

2D machine vision (TDI line scan imaging)
TDI line scan cameras expose the same line on the object multiple times to enable either higher speeds with the same lighting, or the same speed with less bright lighting requirements. TDI (time delay integration) achieves this by using multiple lines in the sensor, which each line exposing multiple times with the exposure or line rate perfectly synchronized to the movement of the conveyor using an encoder. Emergent offers 10GigE, 25GigE, and 100GigE TDI line scan cameras based on the Gpixel GLT5009BSI sensor, which speeds ranging from 121KHz (10GigE) to 608KHz (100GigE).

3D machine vision
Several methods exist for acquiring 3D images, including stereo vision systems, laser line triangulation, Time of Flight, and fringe pattern projection. 3D machine vision is particularly useful in the analysis of complex parts or finished products, as the 3D image can be compared against a digital twin.

Ultraviolet
To reveal defects in materials that cannot be seen in the visible spectrum, some machine vision solutions use high-speed ultraviolet cameras. Use cases include quality control for plastics, glass, semiconductors, and LCDs. Emergent offers several cameras models equipped with Sony’s 8.1 MP UV Pregius S IMX487 CMOS image sensor, which offers increased sensitivity in the UV waveband from 200 nm to 400 nm. This includes the 25GigE Bolt HB-8000-SB-U and the 10GigE HR-8000-SB-U cameras.

Spectral imaging
Multispectral and hyperspectral imaging techniques can capture information from beyond the visible spectrum, especially appropriate for applications such as food inspection and agricultural imaging.

Infrared
Infrared imaging detects heat and spots defects in metal parts and objects that would otherwise go unnoticed.

In the realm of manufacturing, there are innumerable potential applications for process optimization and continuous process improvement with machine vision. Here are just a few examples:

Label inspection
Most consumer products require labels to be printed and affixed properly to packaging. Machine vision provides a way to automize many elements of label inspection, from recognizing when labels are applied upside-down or in the wrong place to ensuring the label does not have blemishes or marks that could make it unreadable. Many label inspection solutions rely on Optical Character Recognition (OCR), the software’s ability to identify text and match it against predetermined values.

Inventory monitoring
Machine vision cameras mounted within warehouses or attached to drones can be used to monitor inventories, saving significant time and effort. Drone-based solutions offer the potential for dramatic cost and labor savings and can be integrated with a company’s existing warehouse management system (WMS) to provide direct information and recommendations to human decision-makers.

Equipment maintenance
Machine vision can be used effectively for equipment monitoring and maintenance. For example, cameras capture images of factory equipment, allowing software to monitor even small movements and vibrations that could cause components to become misaligned. Managers can then use the visualizations to identify and measure improper movement and diagnose faulty equipment.

When evaluating machine vision cameras for process optimization, start with the intended use case to determine needs around resolution, speed, and interoperability with existing equipment, such as PLCs.

These are some major considerations:

EM radiation sensitivity
Cameras range from monochromatic to full color to infrared and ultraviolet. Depending on the use case and type of data, different cameras will be applicable.

Resolution
Available in a range from 0.5 MP to 8KX16, camera resolution needs depend on how detailed the inspection needs to be for the use case in question. Minimum field of view (FOV) and feature size both impact the camera resolution selection and lens choices. For example, inspecting microchips will require a camera lens with a much smaller FOV than if you’re inspecting car bodies in an automotive plant.

Frames per second (FPS)
Frames per second is all about speed. For a machine vision solution monitoring a conveyor belt that processes many items per second, the FPS needs to be very high. Other use cases, such as machine maintenance, may not require high-speed cameras.

Interface
Interoperability can be a challenge for manufacturers implementing machine vision, as some software only works with certain camera brands and models. Manufacturers should look for cameras that are compatible with their intended software and existing factory infrastructure, such as PLCs.

Pixel size
Pixels are the light-capturing elements that comprise any sensor. Most machine vision cameras have pixels ranging from 1.85μm to 9μm, with the larger pixel sizes providing enhanced signal to noise ratios and saturation.

Emergent leverages GigE Vision and the ubiquitous Ethernet infrastructure for reliable and robust data acquisition and transfer, with best-in-class performance, instead of using proprietary or point-to-point interfaces and image acquisition boards. Emergent deploys an optimized GigE Vision implementation with support for direct transfer technologies such as NVIDIA’s GPUDirect, which enables the transfer of images directly to GPU memory. Instead of placing the burden or large data transfers on the system CPU and memory, GPUDirect technology allows the GPU to handle data processing tasks while maintaining compatibility with the GigE Vision standard and interoperability with compliant software and peripherals.

Depending on the type of processes and materials that need to be observed, camera requirements can vary. Emergent Vision Technologies offers an extensive array of camera options to suit disparate imaging needs.

For additional camera options, check out our interactive system designer tool.