- [Alana] Manufacturing and industrial companies typically face one big issue--maintaining their equipment. Just like with any hardware, you're either reacting to maintenance issues, or you're proactively trying to prevent the equipment from breaking. Both of these approaches have their perks and their quirks. Reacting to equipment failures often leads to downtime, and can create significant costs if you need specialized parts or skillsets to fix the equipment. Preventative or predictive maintenance can also be costly, and it may not even fix the problem. It could just result in over-maintenance, or if it's not done correctly or frequently enough, it may fail to prevent the failure from occurring. Doing predictive maintenance correctly requires you to choose the right sensors, connect them together with an IoT gateway, test the system after it's complete, and then eventually transfer the data somewhere to process it. Then, data scientists and ML engineers could come in to analyze the data and create an alerting system for equipment failures. This solution might take months to implement, require a specialized skillset that goes beyond simple data analytics and data modeling, and, ultimately, may be ineffective. It may require even more time to refine and train the solution for it to perform as expected. This is why so few manufacturing and industrial companies have taken the time to implement predictive maintenance for their equipment. But things have changed in the past few years, where cloud companies, and others, have started to offer services that no longer require you to create your own homegrown solution. For example, AWS has a range of industrial ML services that would perform exactly the same functions as the solution we just talked about, with less need for specialized skillsets or sensors. In fact, if you didn't already have an existing sensor network, you could use Amazon Monitron, a service that offers an end-to-end monitoring system, with sensors that capture vibration and temperature data, a gateway that aggregates and transfers data to AWS, and an ML service to detect anomalies and predict maintenance. This could come in handy in multiple ways. Let's take an example. Let's say we want to monitor cooling fans in your factory and predict their maintenance. By using Amazon Monitron's mobile app, technicians can receive an alert that a fan is acting abnormally. The technician can then go to the fan in question, fix it themselves, or schedule maintenance. However, if the fan is still operating correctly, and the alert was inaccurate, then the technician can provide feedback on the accuracy of the alert, and Amazon Monitron will use that to improve over time. The most important part is you can get up and running in a matter of hours, not months, and use the ML model without ML or cloud experience. Although, commonly, companies already have an existing sensor network, but don't have the skillsets to build their own ML models. In that case, the complete end-to-end solution that Amazon Monitron provides isn't really needed. And instead, you can use another service, called Amazon Lookout for Equipment. This allows you to use your existing sensors and send that data to AWS. Lookout will then use this data to create a customized model for your environment that will help predict abnormal equipment conditions. No ML experience and little cloud experience is required. Now, there is another way to monitor equipment that's become very popular, and that's by using computer vision to automate monitoring and visual inspection tasks. For example, let's say you work in a manufacturing setting, and a supplier sends over a set of 3,000 machine parts, and you need to detect defects in those parts. So, after the parts are flagged for inspection, they go through an incoming quality-assurance process. Now, this process can either be completed by a person or by machine. However, it's likely that the person will only inspect a sample of these machine parts, maybe 100, to determine the defect rate. The machine however, can do 100 percent inspection to determine the defect rates for all 3,000 machine parts. That way, you have an exact defect rate instead of an estimated one, and you can send defective products back to the supplier. AWS has a service that can help you achieve that called Amazon Lookout for Vision. It's an ML service that helps identify visual defects at scale, and decrease dependency on manual inspection, in general. This helps us automate the quality-inspection process in your manufacturing lines. Now there are more industrial ML services that I'll link in the readings, but Amazon Monitron, Amazon Lookout for Equipment, and Amazon Lookout for Vision are the big focus for this course. Thanks for staying with us. See you next time.
