※ 기상링크 :
Industry Insights: Guide to AI Software | Association for Advancing Automation (automate.org)
POSTED 04/15/2024 | By: Carrine Greason, A3 Contributing Editor
AI software is solving challenging problems for automation engineers. As more companies look to AI, here is what they need to know.
Artificial intelligence (AI) software helps manufacturers and other industrial organizations wring out micro-stops in production lines, enable robotic systems to adapt to changing circumstances without programming, and alert engineers to emerging quality problems before the first defective part reaches the end of the line.
In this guide, first in a series from the Association for Advancing Automation that introduces AI software, AI middleware and AI hardware, you will learn what AI software is and what you need to know about buying AI software. You also learn about recent use cases that have delivered timely return on investment. To compile this guide, we spoke with industrial AI experts from Acerta Analytics, CMES Robotics, Invisible AI, and ProMANAGE Smart Manufacturing Solutions.
What is AI software?
While automation has long provided ample benefits in industrial environments, smart systems enabled by AI software do far more, because AI software handles what traditional automation cannot—ambiguity. Using data gathered from a vast array of sensors and data feeds in real-time, AI software collects data, recognizes patterns and deviations from the norm, makes predictions, reports results, and triggers action.
AI software makes informed predictions based on massive amounts of data, far more data than people have time to sort through. With these predictions, AI software can take action, such as to separate bad parts from good ones, alert service personnel to the need for equipment maintenance, and use visual intelligence to identify when a worker is performing a task incorrectly. It can then provide them with learning support to get back on track.
Four things to know when buying AI software
When purchasing AI software, industrial businesses need to know four things from the outset:
- AI software is easily understandable: As an industrial buyer, you need to understand what any software you buy or license will do for you, including AI software. Once you decide on the business problem you want to solve, make sure the AI solution solves the problem, including your specific edge cases and whether the solution supports the scale and level of data protection your company requires.
- AI software often requires large training data sets: This affects time to value. Some AI software works out of the box while other AI software requires a training period before it provides value. Training data is readily available in some use cases, such as interpreting human movement and identifying rectangular shipping boxes and most automotive parts. Data is not readily available for other use cases that require visual insight into odd-shaped and bulky items or machine failures that occur infrequently. In cases such as these, the AI software may apply streaming analytics and learn from signals, images and video collected in real time on site.
- AI software makes abundant use of statistics: This can help demystify what the software does. Many machine-learning (ML) techniques are based on statistics as well as other mathematical disciplines. Statistics are key to predictive analytics, which applies automation to the analysis of production data and packages it in algorithms or AI models. An AI model learns the properties of the data and how data from one process relates to data from other processes, which is immensely helpful insight for decision-making. The outcome is a prediction, or inference. Since good decision-making depends on good data, it’s important when buying an AI software solution to know where the raw data comes from that feeds the AI model, and where the AI results from the AI model go to be acted on.
- The software can pay off in one to two years, or in some cases, much faster: AI software pays off for industrial users by delivering concrete improvements in the monitoring and training of workers, detecting process problems that otherwise lead to scrap and rework, predicting the best timing for equipment maintenance, and expanding the types of tasks that mobile autonomous robots (AMRs) can perform. AI software also supports improved data-driven business decision-making for added productivity and can eliminate the need to have experts in robotics, data science, or ML/AI at every site.
Smart use cases for AI software ROI
Many of the industrial projects that provide a prompt return on investment employ AI technologies, such as AI vision, predictive quality, predictive maintenance and smart robots. The following are accounts of recent AI software projects delivered or enabled by Invisible AI, Acerta Analytics, ProMANAGE Smart Manufacturing Solutions, and CMES Robotics.
AI vision systems enable process optimization and improved operator training
Smart vision at scale is a big part of a recent maturation of AI software. AI-based computer vision enables an automated system to interpret the people, machines, and parts in its environment in real time. Visibility enables process optimization and improves safety. This is the type of AI software that Invisible AI delivered to a tier 1 automotive parts supplier, resulting in significant improvements to process efficiency and worker safety. “You need to see to monitor and identify worker safety issues and human ergonomics,” explains Prateek Sachdeva, co-founder and chief product officer at Invisible AI. Quickly interpreting camera and video data is where edge-based AI vision systems shine.
As Sachdeva explains, the sheer volume of information contained in a video requires massive compute power to interpret. But sending so much data to the cloud for processing creates delays and incurs bandwidth cost. “Training smart camera systems at the edge—in a factory, for example—enables real-time AI vision at scale and simplifies deployment of tens or hundreds of AI vision systems as needed, while keeping the video data on premises,” he says. “The Invisible AI visual intelligence platform provides continuous monitoring of stations on a production line—capturing and analyzing video from manual assembly work in real time and intervening to guide the worker on how to perform a task as designed—to ensure work is performed correctly every time.”
As Sachdeva further explains, Invisible AI has trained its AI vision systems to recognize seventeen joints on the human body for more accurate identification of when a worker’s movements differ from best practice. The result? Within just a few hours after deployment of an Invisible AI vision system at a single station at the automotive parts facility, the supplier learned of an opportunity to reduce the steps walked at a station by providing a tool belt pouch, which improved efficiency.
For a similar project at an automobile manufacturer, Invisible AI deployed an AI vision system that revealed a worker was not using the provided hoist to pick up and place a heavy part on an automated guided vehicle (AGV), which put the worker at risk of injury. With the information provided by Invisible AI, engineers knew to investigate.
AI software reduces work through predictive quality and predictive maintenance
The major advantage of using AI models is their ability to sift automatically through mountains of data, looking for relationships across any number of processes, systems and parameters. From the data, AI-enabled software can extract valuable insights, alert engineers and trigger intervention to minimize product defects and equipment failure.
As Greta Cutulenco, CEO of Acerta Analytics explains, predictive quality is the next phase of maturity in Industry 4.0 facilities—moving beyond the machines and to the heart of the parts and products that the manufacturer produces. The software links process, product, and test data, enabling it to immediately find relationships between data signals that influence quality outcomes. It can also accelerate root cause analysis, saving hours, days or even weeks of investigation for quality teams.
In precision manufacturing, parts must meet specifications with minimal variance—which is difficult and costly. Acerta Analytics helps automotive manufacturers use the growing volume of production data to achieve high targets for quality and optimize process efficiency. The company’s predictive quality platform uses real-time process and product data collected across production machines and product lines to automatically identify early indicators of quality problems through ML and AI, Cutulenco says.
Data is ingested as it becomes available at each stage of the production process, including at quality gates, such as end-of-line testing. By immediately processing the data with AI and real-time analytics, the software identifies anomalies and trends much faster, which enables prompt and precise intervention from engineers to maintain part quality.
Like a robot automates a set of steps, AI software automates the analysis of data at scale—each aspect of an injection molding machine has hundreds of parameters, for example. “AI can determine if any of those parameters are changing over time and predict at what point the trend will cause a part to vary beyond tolerance, before a defect even occurs,” Cutulenco explains.
Similarly, predictive maintenance uses AI to analyze data from equipment to infer when and why a machine will fail. Good accuracy in predicting machine failures enables timely service and less downtime. AI models predict maintenance with 90-95% accuracy, says Mustafa Can Ozturk, chief operating officer (COO) of ProMANAGE Smart Manufacturing Solutions, a technology company that specializes in industrial IoT (IIoT), augmented reality (AR), and AI for manufacturing operations.
Machine failure is particularly costly when it slows production and damages equipment. “AI models predict a negative outcome, days in advance, so an experienced service person can get in there and fix the problem, such as high vibration levels, before a machine failure stops production,” Can Ozturk says. After identifying an upcoming need for service, the ProMANAGE solution creates and sends a maintenance order, informing a service technician.
Smart robots use cameras and AI software for autonomous motion control
AI vision boosts the capabilities of AI software and enables autonomous control of industrial robots. For example, tasks that require visual intelligence, such as stacking boxes of various shapes and sizes on a pallet—the industrial equivalent of playing the personal computer game Tetris—can now be performed by robots using AI vision for pick-pack-pallet logistics.
“Traditionally, robotic automation involved teaching a robot to move from point A to point B, which works fine for a task such as stacking equal-sized boxes but not a mixed pallet,” explains Alex Choe, president of CMES Robotics, which offers a full AI software technology stack for robot control.
“With AI, the robot can perceive the world through AI vision and make decisions based on what they are taught,” Choe continues. With the company’s AI software, robots quickly learn to load and unload full and partial pallets of any configuration of boxes. CMES Robotics has ready-to-use trained automation systems that recognize rectangular boxes of every color, bottles used in beverage factories and most types of automotive parts.
As Choe explains, smart robots that can see are able to fill gaps in the labor supply for jobs that workers don’t want, such as placing frozen meat on a conveyer belt in a cold room, or jobs that are unsafe for humans. “We worked with a food manufacturer that had no automation previously but couldn’t find workers to do this chilly job. Yet traditional robotics systems were unable to handle the real-world boxes,” he says. AI software expands the range of boxes that a robot can handle—and robots don’t mind working in the cold, creating yet another of the many industrial applications for AI software.