In an earlier post, IBM industry expert Scott Campbell talked about how manufacturers are pursuing resiliency and Zero D to stop defects and improve products and service quality. In part two of our discussion, he discusses how mitigating rework can save millions and offers some insights on the value of creating citizen data scientists.
Can you explain the concept of “detect and correct” and the kind of technologies and processes you need to implement to reach that kind of efficiency and reliability?
The idea is if you can detect an issue or defect at the point of installation — using AI computer vision models — then you can correct that defect immediately without it becoming cemented into rework. The example I always use is a dashboard of a vehicle: the average vehicle has over 300 electrical connectors, and many of them reside within the dashboard. These have to be manually connected, because they’re wires and not easily managed by machinery. If a connector is not seated correctly, it’s going to short or it’s going to fail. This means that function won’t work. But if you catch the error at the point of installation — and this is where computer vision models are so important — you can determine if it is it fully connected, partially connected, or if the line technician forgot to connect it altogether.
This detection capability can also be integrated into an overall quality system and/or enterprise asset management system. In the case of Maximo Visual Inspection, it is tightly integrated into Maximo Application Suite for enterprise asset management and performance, while easily integrating into customer quality alert systems. So, when a defect is detected, it can immediately signal an alert on the manufacturing floor to ensure the worker verifies and fixes the issue before it moves to the next assembly process. This immediate alerting is what avoids expensive rework. In the case of connectors within the dashboard, if defects go undetected, the rework fix for a simple connection gets exponentially more expensive, as it often requires dashboard removal and re-installation.
Using computer vision and AI to see the errors before they turn into rework and fix them right then — and in some cases a company is willing to stop the manufacturing line to fix a problem before it gets cemented — is a pretty significant capability. Especially because scrap and defects can cost a company more than 10% of annual revenue.
When people think of AI models, they think of data scientists and the difficulty in hiring expensive resources who understand AI technologies, deep learning neural networks and specialized AI computer vision models. But what IBM has done is made it extremely easy for the subject matter experts (SMEs) — the people that know what they’re looking for defect-wise — to actually create and manage the AI models. We do it through a user interface that requires no code.
It’s literally labeling a few images within a picture as good or bad or any other decision criteria they wish to define. Then the system can provide auto-labeling based on what has been labeled thus far, greatly reducing the workload. Finally, the existing data set can be augmented to create very large data sets out of the original sample size. This provides the data to build models that give predictable outcomes — in most cases, the accuracy is high as 95% to 98%. The result: subject matter experts take control of the actual models without the need for data scientists. This makes adoption a lot faster because companies use the people who are familiar with what the manufacturer is doing on the assembly line. That expertise is also a major contributor to the high-level accuracy of the AI models.
What about the concept of predict and correct? Does that play a role in driving continuous operations?
At IBM, we asked, what if you could increase efficiency, extend asset lifecycles, reduce downtime and costs — all while building resiliency and sustainability into your business?
Predict and correct is fundamental to being able to answer that question.
We’ve made it easier to digitize operating environments by taking the sensor data coming off of assets, and understand at a point in time the condition and operational status of those assets. And it’s a lot of data! A single production line can produce more than 70 terabytes of data each day.
By understanding the asset’s total health in terms of lifecycle and leveraging historical time series data, Maximo can predict when a failure is likely to occur in the future. If you can accurately predict failure well before it happens, you can remediate it. This predict and correct capability plays a major role in delivering and facilitating continuous operations.
You start with Maximo Monitor — capturing data and gaining visibility into what your assets are actually doing. Then you add Maximo Health, which tells you from a lifecycle perspective what maintenance structure you should be looking at and allows a single view of assets across the enterprise. Finally, with Maximo Predict, you can see into the future to be much more prescriptive with your asset performance management. It’s an evolution, but Predict is where the AI models come together to allow a customer to see where there is probability for failure for all of their assets and take corrective action.
We’ve been talking about the auto industry but I’m assuming that any industry can benefit from this.
Absolutely. And it bridges beyond manufacturing. We’re talking about the pursuit of Zero D and resiliency for manufacturing because it aligns so well to Industry 4.0, but the same technology can be used, for example, in travel and transportation. Consider railways and the ability to understand the assets — which are both the railway tracks and the train itself — and looking for potential failure. Sensor data is part of it, but then AI visual inspection can also be used to visually inspect railcars, wear on couplings, wheels, and wiring as just a few examples.
Traditionally, with cargo trains, there are maintenance yards, and the train will pull in and then maintenance people manually inspect the train. They visually ensure everything is okay before they let it go back on the track. But that industry is quickly evolving to provide inspection while the train is in transit. Cameras over the tracks take pictures of the train and provide the results immediately through AI computer vision models. If there are urgent safety concerns for example, the railway operator could stop a train. If not, it could continue on, but the technicians might say, Okay, the next maintenance window we’re going to need to make these repairs. Not only is the inspection much more complete — and can happen with higher frequency — but it is also much more accurate in prediction, because it’s using sensor data as well as visual data to manage the assets. We’re also seeing this in civil infrastructure and with bridges and roadways. There’s just a lot of places where visual data and sensor data come together.
What are some of the issues and misconceptions that an organization might have when it comes to using AI to predict asset health and build a more resilient organization?
From a challenges perspective, the first one is a company that doesn’t use IBM Maximo EAM (Enterprise Asset Management) as its work order system. Often companies believe they can’t take advantage of the rest of our application suite if they don’t use Maximo EAM across their entire organizations. But IBM’s monitor, health and predict solutions can connect to other EAM systems so that companies can take advantage of their operational data. We can also connect to other systems that are gathering the sensor data and we can feed this data into Maximo Monitor. This is important because two-thirds of operational data goes unused. It also eliminates a hurdle a company might have to jump with another provider, simply because their work order system is in another vendor’s application. We can manage within that and still drive value with predictive capability by introducing monitor, health, and predict.
Another typical issue is that each asset has siloed data into its own repository. Getting data across all the assets collected into a single pool can be very difficult and time-consuming. But we can bring connectors via APIs or solutions like IBM App Connect and help customers consolidate data into a single repository. This repository can capture time series data, and then you have a starting point for building resiliency and sustainability into your business by extending asset lifecycles and reducing downtime and costs. Once you’re positioned for intelligent asset management — and building resiliency and sustainability into your business — you can reduce operational costs up to 25% and increase uptime and availability by 20%. Those are results that no one objects to.
The post Intelligent asset management and the race to Zero D appeared first on IBM Business Operations Blog.