This blog post introduces a new novel product solution from SAP Digital Manufacturing Cloud to integrate machine learning into business processes for visual inspection. A series of blog posts will follow to show how to train a new machine learning model and integrate this model to be used to assist the operator on the shop floor, all within some minutes instead of what takes usually months and without a deep knowledge for machine learning. Also we will explorer soon other extensibility options to implement your own AI/ML solution for Visual Inspection.
Despite enormous progress in Artificial Intelligence, one of the biggest challenges remains how to operationalize machine learning for end-to-end business processes. Today, a wide variety of machine learning tools are available for data scientists, but also for business users who have limited knowledge how to train machine learning. It has never been easier as it is today to train a machine learning model within minutes providing reasonable results.
The challenge comes once the machine learning model is ready and needs to be integrated into the business process where it is intended to provide value by assisting humans or even automate process steps. This integration can take weeks to month to be implemented as it touches business-critical systems and requires scalable technology. Most machine learning models never really are used productively, because of this hurtle, typically referred to as the machine learning production gap.
This integration into productive systems becomes even more challenging in certain environments, such as manufacturing industries. On the shop floor, humans, machines, robots and IT systems ensure the production is running and products are manufactured at the right time, in the right amount, and in the required quality.
Take Visual inspection as one example:
Visual inspection is one of the critical processes ensuring the quality of the manufactured product and can be found throughout the value chain: starting in the warehouse where goods are received. Following the value stream, visual inspection varies from worker self-checks, acceptance testing, to fully automated machine visions systems. Visual Inspection is even relevant after products have left the plant and are shipped to customers: field technicians inspecting the assets have to visually assess if certain maintenance tasks are required or to identify the right spare parts based on computer vision.
Computer vision is one of the most advanced examples of AI with applications in medicine, autonomous driving and surveillance, outperforming human experts in accuracy, speed and endurance. Typically, deep neural networks are trained on millions of images of various classes. In a manufacturing context, such a trained model can be reused to train another model using only very few images of defective and good parts, which allows visual inspection applications on the shop floor.
With the latest releases of SAP Digital Manufacturing Cloud, the new AI/ML scenario Visual Inspection was released to assist the operator in identifying defective parts and logging nonconformance to ensure that defective parts are automatically redirected to the repair work center or to initiate other corrective actions. This empowers the operator on the shop floor to focus on value-added processes instead of performing repetitive tasks which can be mentally exhausting and might introduce human error, leading to undetected quality problems.
In upcoming blog posts, my colleagues involved in the development and realization of this new visual inspection solution will provide more details:
SAP Links:
The production gap in Artificial Intelligence
Despite enormous progress in Artificial Intelligence, one of the biggest challenges remains how to operationalize machine learning for end-to-end business processes. Today, a wide variety of machine learning tools are available for data scientists, but also for business users who have limited knowledge how to train machine learning. It has never been easier as it is today to train a machine learning model within minutes providing reasonable results.
The challenge comes once the machine learning model is ready and needs to be integrated into the business process where it is intended to provide value by assisting humans or even automate process steps. This integration can take weeks to month to be implemented as it touches business-critical systems and requires scalable technology. Most machine learning models never really are used productively, because of this hurtle, typically referred to as the machine learning production gap.
This integration into productive systems becomes even more challenging in certain environments, such as manufacturing industries. On the shop floor, humans, machines, robots and IT systems ensure the production is running and products are manufactured at the right time, in the right amount, and in the required quality.
Take Visual inspection as one example:
Visual inspection business critical process throughout the value chain
Inspection by human or machine using vision can be found throughout the value chain
Visual inspection is one of the critical processes ensuring the quality of the manufactured product and can be found throughout the value chain: starting in the warehouse where goods are received. Following the value stream, visual inspection varies from worker self-checks, acceptance testing, to fully automated machine visions systems. Visual Inspection is even relevant after products have left the plant and are shipped to customers: field technicians inspecting the assets have to visually assess if certain maintenance tasks are required or to identify the right spare parts based on computer vision.
Computer Vision is ready to use
Computer vision is one of the most advanced examples of AI with applications in medicine, autonomous driving and surveillance, outperforming human experts in accuracy, speed and endurance. Typically, deep neural networks are trained on millions of images of various classes. In a manufacturing context, such a trained model can be reused to train another model using only very few images of defective and good parts, which allows visual inspection applications on the shop floor.
Fully integrated machine learning to assist operator on the shop floor
Operator view on Production Operator Dashboard executing a visual inspection, assisted by machine learning
With the latest releases of SAP Digital Manufacturing Cloud, the new AI/ML scenario Visual Inspection was released to assist the operator in identifying defective parts and logging nonconformance to ensure that defective parts are automatically redirected to the repair work center or to initiate other corrective actions. This empowers the operator on the shop floor to focus on value-added processes instead of performing repetitive tasks which can be mentally exhausting and might introduce human error, leading to undetected quality problems.
One of the key focuses of the new AI/ML scenario for visual inspection was to simplify the process to integrate a new machine learning model into the business process within minutes instead of months. At the same time, it supports the use of the latest machine learning technology for computer vision as this is still an area where new deep learning models are developed and released to the public, almost monthly, which outperform the previous best benchmarked computer vision solution. To address both challenges at the same time, Visual Inspection in SAP Digital Manufacturing Cloud allows users to upload a trained machine learning model (also referred as Bring Your Own Model) and activate it for use on the shop floor without writing a single line of code. As a positive side-effect, Visual Inspection runs on the edge which means the model runs locally on the shop floor, not requiring the images to be send to the cloud for analysis, allowing robust and resilient predictions within sub-seconds.
In upcoming blog posts, my colleagues involved in the development and realization of this new visual inspection solution will provide more details:
- A quick tour shows in few steps how within a few minutes a business user can train a model, upload the model using a simple wizard, and use the machine model to assist the operator on the shop floor executing the visual inspection task.
- The range of possibilities to train a machine learning model for visual inspection and prepare it to be used in SAP Digital Manufacturing Cloud from perspective of a data scientist.
- How to setup SAP Digital Manufacturing Cloud, starting with the master configuration of a plant, publish a Production Operator Dashboard for Visual Inspection and use industrial cameras to capture image for inspection.
- Another blog post is planned to show how to use extensibility options to integrate other visual inspection solution with SAP Digital Manufacturing Cloud, This enables customers, partners and startups to extend the standard solutions which can address manufacturing specific problems for different industry use cases.
SAP Links:
- Please check out the roadmap explorer for latest update on delivered and planned features in the area of AI and ML at SAP Digital Manufacturing Cloud
- For more insights about the capabilities of SAP Digital Manufacturing Cloud please check out the product overview
- Ensure to follow the tags Artificial Intelligence and Machine Learning to receive updates about the planned blog posts. Visit the SAP Community topic page for AI and ML for more information.
- SAP Managed Tags:
