content provided by pleora

Artificial intelligence (AI) has received a lot of attention over recent years. It has also brought awareness to traditional machine vision processes in manufacturing and industrial automation. Yet, there is one aspect that is often not discussed in these broad sweeping articles – AI capability against risk.

According to a recent report from McKinsey, companies that initially jumped at the “hype of AI” are now understanding what it can and can’t do. Manufacturers are now identifying areas where they think AI can provide an attainable return on investment (ROI) within visual inspection processes. We’ve always been told that given the data, deep learning can solve any challenge.

However, for the manufacturing industry the data required for quality inspection and defect detection can be difficult to cultivate. To address this challenge, manufacturers are discovering a sweet spot in combining traditional machine vision approaches with AI capabilities.

Digital & Augmented Work Instructions as Apps 

Manufacturers who have heavily relied on paper-based instructions are showing an increasing interest in digital work instructions. Digitizing work instructions can support an operator by providing more detailed information, such as notes, images, and videos. This base level digitization goal is to provide the user with short, concise, and clear understandable step-by-step instructions with supporting information and additional details available if required. If manufacturers would like to take the concept of digital work instructions to the next level, they can add augmented reality using machine vision.

Like a virtual reality (VR) set, an operator is guided step-by-step through their work instructions using the camera-based system. The user carries out their tasks while viewing their actions through a display. The display reflects the operator’s actions in real-time, with augmented reality providing notes, instructions, or warnings. In addition to supporting the operator with augmented work instructions, the system can layer AI and machine vision checks into the process to ensure proper quality standards are followed. Manufacturers benefit from initiating a quality check in parallel with assembly processes. This additional layer adds efficiency and higher quality standards into the process at the beginning, for end-to-end quality inspection.

1. Inline Product Inspection 

Product inspection is one of the most common areas for AI and machine vision applications. Manufacturers and quality managers often find themselves evaluating how automation can fit into quality checks for real-time inline inspection. Quality inspection is a main pillar of the traditional machine vision space. This industry requires flexible solutions that can be applied across a wide range of inspection applications for products covering virtually every market. For many machine vision applications a model to classify or locate defects may require algorithms from open source libraries, like OpenCV or TensorFlow. However, these will only work when supplied with an abundance of data.

In this case, the best approach is to combine AI and traditional vision as a hybrid approach, with traditional vision preprocessing images and identifying defects and AI using data to reduce false positives over time. This hybrid approach can be implemented easily by simply adding AI to an existing inspection system. Common challenges with inline inspection:

  • Time: There can be substantial time requirements prior to observing a ROI, especially for pure AI projects that require a lot of collected data and manual labelling.
  • Cost: Depending on how quickly a production line is running, expensive hardware and processing power may affect cost. On the inference side, if an AI models is not optimized for power efficient edge platforms, it can cause limitations with scalability.
  • False positives: Despite advances in the field, depending on the application you may still not get the accuracies required and human intervention may still be required for secondary inspections.

2. Assisted Quality Control (QC) Offline Inspection 

Just like with robotics, where more difficult tasks are not easily automated and the solution brings forth the concept of “cobots” (robots co-working with operators), the same applies in vision applications with offline inspection systems. The main goal with most offline inspection systems is to provide decision-support that helps operators identify defects. In print and packaging applications, this includes tasks like comparing for content errors (incorrect images, text, barcodes, braille, etc.) against a master golden reference image.

In addition to offline inspection, this type of check can happen at many different points in the process, like pre-press, supplier proof checks, print setup and incoming quality control (IQC). In printed circuit board (PCB) and electronics manufacturing or assembly, automated optical inspection (AOI) is widely used to perform batch checks for surface mount technology (SMT) components. As boards are handled by different people through the manufacturing process, there is a potential for damage. A final offline inspection is ideal before boards are packed and shipped.

Secondary inspections, where manual visual inspections are performed after the products are “filtered” through an inline inspection system, can also be a time-consuming task. By adding AI, the operator can train the system in parallel with their decision-making so AI can provide an automatic recommendation in the future. The benefit to this process is an almost immediate ROI, where a manual task is now significantly improved without having to disrupt the existing production processor or spending additional training time.

3. Factory Monitoring and Surveillance 

Health and safety are another opportunity for AI automation. In place of basic surveillance cameras used to monitor a production line or a restricted zone, AI deep learning models can provide both automated monitoring and data collection as part of a factory analytics program. Manufacturers can then use the data to track operations, including providing immediate alerts to supervisors if a worker is injured or if there is unauthorized entry.

4. Predictive Maintenance and Analytics

Predictive maintenance and analytics can also be improved with the use of AI in vision applications. Quality managers can detect upcoming maintenance requirements using a combination of sensors and cameras that analyze the performance and health of equipment. This equipment may be the same production equipment used for quality assurance, or dedicated test equipment. AI can provide analytics from sensor data and cumulating trends, alerting operators before equipment needs repair or calibration, saving facilities in line down time, and unexpected costs. AI is enabling a great opportunity for deeper insights into your production process. Quality control can use AI data for a better insight into the number of defects produced in the facility, including subcategorizing the data by type or criticality. This will build visibility on where major errors are happening in the manufacturing process, allowing root cause analysis to be a much simpler and cost-effective process. This type of predictive analytics also opens the door for Industrial Internet of Things (IIoT), which allows data to be easily exchanged between systems using protocols like MQTT and OPC-UA to build standard Industry 4.0 dashboards.

How can I get started with AI and Machine Vision? 

Get started with a pilot project that doesn’t interrupt your current production workflow. With an offline application, users can automate or augment a manual task like work instructions, image comparison or object counting. From there, AI can capture data over time to help train its model, eventually preparing itself to transition over to inline inspection.