5 Ways Industrial AI is Revolutionizing Manufacturing
Artificial Intelligence (AI) is most commonly applied in manufacturing to improve overall equipment efficiency (OEE) and first-pass yield in production. Over time, manufacturers can use AI to increase uptime, improve quality and consistency, which allows for better forecasting.
As with many components of digitization, AI implementation can seem overwhelming. Concerns about how to effectively use and manage billions of data points that are now generated by intuitive computing power and their connected machines are common amongst manufacturers. Many are uncertain how to get started and often attribute their caution in AI adoption to cost, IT requirements and/or fear of not being “Industry 4.0” ready.
In order to stay competitive, it’s important manufacturers adapt to a more data driven business model. This often includes reorganization of staff, hardware and software upgrades.
AI, a concept often associated with the future, is now a reality and can be applied to your factory today.
Here are 5 Ways Industrial AI is Revolutionizing Manufacturing and tips on implementation :
- Predictive & Preventative Maintenance
Some of the biggest down times for a production operation can be caused by a core piece of machinery being offline due to some type of mechanical or electrical failure. Usually the failure can be easily prevented by following up on the machines recommended preventative maintenance schedule. Often PMs are overlooked or not optimized for the best timeline to complete. Now with the power of IoT devices, sensors, MES data and machine learning algorithms manufacturers can utilize many machine data points to predict breakdowns. PM schedules can be optimized prior to the predicted breakdown to keep machines in top notch condition and the production floor running smoothly.
- Supply Chain Optimization
Today's supply chains are super complex networks to manage, with thousands of parts and hundreds of locations. AI is becoming a necessary tool to get products from production to customer in a timely manner. With machine learning algorithms manufacturers can define the optimized supply chain solution for all their products. Questions like ‘How many resistors should be ordered for the next quarter?’ or ‘What's the best shipping route for product A’ can finally be answered without relying on a best guess approximation.
In-house inventory management can be a major challenge in itself. The production line heavily relies on inventory to keep the lines fed and producing products. Each process step requires a certain amount of components to operate, once those are consumed it needs to be replenished in a timely manner to continue processing. Keeping the factory floor stocked of all necessary inventory is a challenge that AI can help manage. AI can look at component quantities, expiration dates, and optimize the distribution throughout the factory floor.
- Production Optimization
Process optimization can be a data heavy task involving countless historical data sets. Pinpointing which process parameters produce the highest product quality is not an easy task. Manufacturing and Quality engineers run dozens of Design of Experiments to optimize process parameters all the time but often they can be costly and time consuming. With the fast data crunching speed of AI, engineers can find the optimized process recipe for different products. Questions like ‘What conveyor speed or temperature should I input for the highest yield?’ or ‘What machine should I use for this high pitch emerging technology circuit board?’. AI will constantly learn from all production data points to continuously improve process parameters.
- Predictive Yield
Yield prediction conversations always come up when AI in manufacturing is being discussed. The ROI on having a high accuracy prediction AI model is limitless. Predicting yield can better prepare supply chain and inventory management on future component needs. Knowing if yield will be lower than expected can alert production management to increase production time to meet demand needs. Yield prediction is a data heavy complex problem that will require AI to solve.
- Augmented and Virtual Reality
With augmented and virtual reality technologies improving everyday, with more major companies developing devices for this market, it's only a matter of time before the manufacturing industry fully adopts their use. Virtual reality can help better train product builders to perform assembly or preventative maintenance tasks. Augmented reality provides real-time reporting driven by machine learning, on the factory floor or in the field, helping to quickly identify defective products and areas of operational improvement. AR/VR manufacturing applications are endless and can play a significant role in solving today's challenges.
- Energy Management
AI can help the often overlooked area of energy management. Most engineers don't have the time to analyze the cost of factory energy consumption. Having an AI look into energy consumption of a production operation can significantly reduce operations costs. Reduced cost can allocate more funding for process improvement resources which can lead to higher yield and quality.
Now that we have reviewed the benefits, let’s take a look at the ease of implementation, functionality and deployment.
Intraratio AI/ML: Implementation Architecture
In order to implement a highly productive, flexibility confident application Intraratio uses Python which is the best fit for machine learning and AI-based projects. There are a lot of great AI / machine learning libraries and frameworks within Python due to its wide community use and popularity of the language.
Our AI application is built on top of the Flask framework, a micro web framework based on Python, to deliver the services as a web-based application. This allows us to deploy on multiple platforms regardless of OS or device, with users able to connect using a browser. Moreover, it brings flexibility and high availability since it can be easily accessible from anywhere.
For the purpose of cross-platform compatibility and simplicity, we choose REST (Representational State Transfer) as the API (Application Programming Interface) for communication with the application. In addition to that, we use OAuth which is the industry-standard protocol for API authentication to ensure a secure and authenticated connection.
Due to the heavy data requirements of an AI base application, we implement a task queue to manage the requests. Each analytics job submitted is enqueued and processed as FIFO (first in, first out) to protect the application from overwhelming requests, and to manage computing resources most efficiently. Moreover, requests are cached so that we can maintain fast performance and eliminate any duplicate requests.
The Secret Sauce
AI/ML requires properly categorized data, with well defined features. A feature is an individual measurable property or characteristic of a phenomenon being observed: either categorical values such as BOM item, supplier, machine model or serial; Or measured values such as time, temperature, thickness, optical intensity, power, etc..
It is absolutely critical to have informative, discriminating and independent features in order to effectively apply AI/ML algorithms of pattern recognition, classification and regression.
Machine and sensor data lack the full suite of values needed to associate with features such as time, job, lot number, part supplier, operator, board serial, component, unit-under-test serial number, measurement data, etc.. Without this categorical data binding, you have no way of tying AI/ML output results back to actual specific product, machine, supplier, etc.
Intraratio’s approach is to automatically bind relevant MES (Manufacturing Execution System) traceability data to the outbound machine or sensor data, in real-time as it is read from the source, and store it fully cataloged as such in a scalable analytics database.
This instantly removes 80% - 90% of the time Engineers and/or Data Scientists lose to cleaning up, manually cataloging, and removing invalid data points, before being able to apply any type of analytics let alone AI/ML algorithms.
Using this approach, the marginal costs and complexity associated with running AI/ML in a production facility become near zero post deployment. This is only feasible if you have a robust MES that can deliver data in real-time, with direct integration with an analytics system, generally the preserve of only fully integrated platforms of which Intraratio is one of the few companies in the world to provide.
What if you had a system in place that automatically detected production issues in real-time, before they happen?
The benefits would be predictive maintenance, inventory and product outlier detection in an accessible and intuitive way, driving operational excellence to new levels.
This would be a game changer to your competitive advantage.
Yes. Data is the new bacon. And AI/ML is taking it to new heights.