Artificial Intelligence (AI) in manufacturing is very likely to be valued at almost USD 16 billion by 2026 (compared to roughly around USD 1 billion being expected in 2020). This is driven by the same factors that push forward the adoption of machine learning and deep learning techniques across industries. These are namely, new algorithms design, specialized hardware and cloud services becoming available, so-called data explosion (Big Data) enabling quality AI training, both open-source and proprietary software libraries development, growing investments, number of applications and an increased demand. In the case of Industry 4.0, however, AI adoption is further driven by a growing importance of general automation, robots and in many cases closely related industrial Internet of Things (IoT) deployments like sensors measuring production parameters, cameras taking snapshots along production lines etc. For instance, in the case of the auto industry, the McKinsey Global Institute indicates that robotics and AI technologies have already advanced to the point where it would be possible to automate at least 30% of activities in about 60% of manufacturing occupations in the United States and Germany.
AI gives machines the abilities once reserved for humans. They can process unstructured data like images, videos, sound, text and find objects, shapes, patterns, trends etc. This can be further turned by AI into information about quality, found incidents, machinery maintenance and even wrapped up into reports by combining these with data extracted from documents or sensors for example. Automation with AI is not just about replacing human jobs though. It is more about doing certain tasks more efficiently and with better quality. In many cases, this becomes a natural result of a productive robot-human collaboration. For instance, it is stated that AI-assisted visual inspection cameras can detect defects with around 90% greater accuracy than humans. Machines like that can obviously do so within a fraction of a second and never get bored. On the flip side, monitoring the quality of products or processes like 24/7 is not a dream job for humans anyway. Therefore, they can focus on more creative jobs like bottlenecks troubleshooting, meeting delivery dates, or taking their smart factories to another level of automation, further improving its efficiency, increasing customer satisfaction, and closing the quality gaps.
It seems that manufacturing is on the brink of another disruption as it gradually embraces the fourth industrial revolution. And in fact, Artificial Intelligence is becoming the key technology changing the way we produce or distribute goods. Forbes, for instance, mentioned some interesting benefits enterprises saw after applying AI in their factories. While you can read the whole article here, let me quote just a couple of examples:
· Caterpillar’s Marine Division is saving $400K per ship per year after machine learning analyzed data on how often hulls should be cleaned for maximum efficiency.
· General Motors analyzes images from cameras mounted on assembly robots, to spot signs and indications of failing robotic components with the help of its supplier. In a pilot test of the system, it detected 72 instances of component failure across 7,000 robots, identifying the problem before it could result in unplanned outages.
One thing humans and machines have in common is that both make mistakes. Therefore, industrial automation brings value as far as it can effectively bring out the best of both worlds. As simple as it might sound, it is not a trivial task at all. We have made a strategic decision at byteLAKE to work with various industry leaders and talented researchers in efforts to combine human knowledge, industry expertise, and know-how with the best AI algorithms. We are creating a product named Cognitive Services, a collection of AI models designed to address Industry 4.0 needs. Each AI model we designed and trained is razor-focused on specific industry tasks, therefore ensuring maximum accuracy. In essence, byteLAKE’s Cognitive Services focus primarily on the following two areas:
· AI-assisted visual inspection for efficient quality of products & process monitoring.
· AI-powered Big Data / IoT sensors data analytics to find trends, enable predictive maintenance, answer questions like why something happens, what will likely happen or to find the collective meaning of the data extracted from many sources.
Humans and machines both make mistakes. Therefore, byteLAKE’s Cognitive Services have been designed to effectively bring out the best of both worlds.
The product is scalable and can efficiently monitor single production lines as well as many distributed across different plants. It can be deployed on constrained, embedded devices (Edge AI, intelligent cameras etc.), on both CPU and GPU powered edge/desktop PCs (both Windows and Linux), in the cloud or in on-premises server installations. It can be easily integrated into factory management and monitoring systems and rapidly enhance their capabilities with AI. It is also possible to leverage Cognitive Services’ modules to jump start your product development.
byteLAKE’s Cognitive Services for AI-assisted visual inspection
These models can be trained to detect and count defects, issues, or goods in production lines or while monitoring conveyor belts. As illustrated above, it can for instance inspect goods traveling on conveyor belts, count them and measure how far they are from critical areas. Besides, the models can be re-trained to visually monitor and inspect other types of things or processes, and again measure quantity and assess quality.
byteLAKE’s AI models for visual inspection will be available in many different versions, tailored and optimized to various industries and their specific scenarios. We will be describing these in more detail on byteLAKE’s blog. Scenarios we are currently working on include:
· Food recognition for efficient self-service and automated billing in restaurants and hotels
· Intelligent cameras for paper mills, detecting issues thru visual analytics of paper production processes (i.e. wet line monitoring)
· Counting goods on conveyor belts, identifying them, and monitoring quality and finding defects
· 3D data analytics for robot arms optimal movements
byteLAKE’s Cognitive Services for AI-powered Big Data / IoT sensors data analytics
The goal is to find answers hidden in Big Data. We are having the following features on the roadmap:
· Improve Client Experience
(analyze past interactions/behaviors to enable personalized decisions)
· Predict outcomes and trends
(detect anomalies or suspicious behaviors, patterns and make recommendations)
· Process Automation
(document processing automation, conversational chatbots, automate complex and repetitive tasks)
· Better Decisions
(Machine Learning, Deep Learning powered)
Currently, the majority of our efforts in that space are focused on the following example scenarios:
· Devices/buildings characteristics analytics based on IoT sensors data to find optimal configurations (i.e. minimize energy consumption)
· Big Data analytics for Agriculture to efficiently assess seeds quality and potential
· Historic data analytics to support agents in pricing negotiations and suggest optimal proposals
How to start with byteLAKE’s Cognitive Services
A typical engagement scenario starts with a 1-hour consultancy (online) during which we discuss the process to be monitored, expectations, data types etc. Then we usually need sample data for initial testing. Next, we guide our clients to prepare proper data sets. Our data scientists are here to clean them and eventually transform them into optimized packages for AI training. Neural network configuration and the actual training follows that step. It is a straightforward process as it leverages our Cognitive Services product. To ensure optimal results we also benchmark the results for various architectures (CPU/FPGA/GPU/Edge) and offer a complete, end-to-end service:
· byteLAKE’s Cognitive Services optimization and training for client’s specific scenario. The goal is to configure the product so that it detects objects, shapes, or analyzes the data according to the scenarios our clients need. In other words, a baseline product can handle a growing number of scenarios. However, we need your input (data) to calibrate it to your needs and i.e. count packages, detect stains on conveyor belts etc. It is a similar process to selecting additional packages for your car. We license a baseline and then we tweak it to your needs: engine variant (=what it shall detect), color (=how the results shall be visualized) etc.
· byteLAKE’s Cognitive Services integration is about deploying our product in your environment. Typical tasks include helping you select cameras or connecting the product to the existing one(s), helping you properly select, position and calibrate cameras, integrating our product with your existing software components, connecting the product to your IoT infrastructure / importing your data, formatting the results etc.
We offer end to end services and bring a solution to a problem:
· Integration and development of (custom) full-stack software. In other words, depending on client needs we can go beyond AI products and deliver full solution: byteLAKE’s AI products integration, backend/frontend/web, cloud/desktop and mobile components integration, design, and development.
· Optimization and guidance about overall architecture and deployment strategies. Thanks to a broad network of our partners, we can benchmark the solution for various deployment needs and deliver both: AI software licensing and hardware components.
For instance, the below illustration depicts byteLAKE’s Cognitive Services integration for manufacturing as part of an end-to-end solution integration with our partner Lenovo Data Center.
Reach out to us to learn more! Contact us at CognitiveServices@byteLAKE.com.
Read more in the related blog post series: Artificial Intelligence for Industry 4.0 (blog post series) — table of contents.