Artificial Intelligence (AI) is one of the most frequently referred items in the news, yet many news reports, TV shows, and movies fail to show what AI really is and why it is so valuable to digital transformations. AI will be a major component of the ongoing digital transformations happening today and will become critical to staying ahead of the competition and growing the business in the years to come.
It Begins with Machine Learning
Before we can examine the impacts of AI on the optimized factory, an important distinction needs to be made between Machine Learning (ML) and AI. Most news articles focused on AI confuse ML with AI. The reality is the two types of technologies are different, but closely related.
ML technologies are fundamentally focused on teaching computer systems what specific data means. For example, consider an automated door in a factory. A sensor on the door will capture each time the door opens and then closes. If the business goal is to identify when the opening mechanism needs maintenance, the opening and closing data needs to be captured as well as the specific data the sensor reports when the door has a problem. The door might open and close 20,000 times, though, before an issue occurs. ML technologies would analyze all 20,000 events to begin understanding normal operations. 1 failure event, though, won’t be enough information to clearly identify when future maintenance actions will be required. The first 20,000 data points start training patterns, but further failures are needed to develop more realistic patterns.
Now consider if the sensor were a video device being used to identify a weakness in the cable being used to operate a crane. Just like the door example above, ML technologies need to be used to train the computer on the difference between a safe cable and an unsafe cable. The complexity, though, increases significantly in this example as the computer needs to be trained based on mathematical representation of pixel data since the collected data are now photos or video content.
In both examples of a sensor on a door and a video device capturing details on the crane cable, the computer needs to process a significant amount of data to build the patterns. This process is often called "cold path analytics." Once the models have been established, the underlying algorithm created is then applied to real-time data using complex event processing tools or "hot path analytics." These tools use the ML output to take immediate action on issues with the door or the safety of the cable.
Moving from ML to AI
AI technologies, then, become the applications and services that apply ML models to produce answers and drive innovations. Companies like Microsoft have spent decades building models, training systems, and creating packaged services that anyone can use in their apps today. These include optical character recognition (OCR), translation services, speech-to-text, natural language input, image recognition, and video intelligence to name a few. Each of these services are individually packaged allowing any developer to bring the capability natively into their custom applications. For example, this is why you can go to a search engine, do an image search for “dog family beach” and immediately see a page full of images of families at the beach with their dog. Image recognition models have been created in ML and then packaged into AI services for what a dog is, what a family is, and what a beach is and then integrated into the search service.
While companies like Microsoft have created more than 30 of these pre-packaged AI services, tools are available to create custom AI apps and services that connect into line of business apps allowing specific business needs and issues to be addressed. Further, composite services can be created that connect both pre-packaged AI services and custom AI services together to address a specific business need.
What Do ML and AI Mean to the Optimized Factory?
For manufacturing organizations that are just getting started with IoT solutions and creating a smart and optimized factory, ML is one of the most important initial steps. One of the most common business scenarios that organizations start with is predictive maintenance. Knowing when a problem might happen and being able to resolve the issue preventing operational downtime can create significant savings and bottom-line improvements. When these business cases are presented, the first question should always be, “Do we have the data?” To create the predictive model required, all the normal operational data is required as well as the anomalous data. Initial ML models can then be created to identify the data patterns ahead of an incident. The initial ML model will be imperfect, but ongoing training and adjustments will ultimately lead to a more consistent and accurate model for issues.
Beyond thinking about specific ML models and data needs, one of the most critical aspects in the design needs to be a vision of the future and the types of services, business models, and operational models that will drive the business. Examples might be:
- Can total yield be increased using the existing resources?
- Defect rates will decrease by 75 percent
- Introducing a custom manufacturing service will drive new customer revenue.
- Creating a new software + services business supporting our products will create new revenue opportunities.
Ultimately AI and ML models can then be created that drive and support these business needs. As an example, ML models can be created that more accurately predict and detect potential defects. These ML models are applied into an AI service that monitors the production. When a potential defect is identified in the process, the AI services can rapidly adjust the manufacturing processes, prevent the product from being delivered to a customer, or even back the process up to automatically fix the issue.
Introducing these advanced services could significantly change how people work in the manufacturing process and roles. AI capabilities will replace many roles and tasks. Those working in the facility will become more focused on the final product, enabling greater levels of automation, and moving into greater levels of service delivery. The employee impact of AI can become significant and will create demands for new types of job roles and responsibilities. Planning not only the ways in which AI will address the operations changes, but also the people changes will become imperative for organizations that want to become or remain the leader.
How Microsoft’s Azure Platform Drives Innovation in ML and AI
Microsoft has invested heavily in AI and ML technologies for many years and Microsoft has been actively integrating AI tools into every single product and service. As a result, organizations have a lot of choices in the types of solutions that will be created. Three of the most important tools from Microsoft for driving innovation in the manufacturing experience are Azure Databricks, the AI Toolkit for Azure IoT Edge, and the Azure Cognitive Services.
- Azure Databricks is a collaborative analytics and ML platform built on top of Spark. What makes Azure Databricks so valuable when considering solutions for the Optimized Factory is the collaborative experience between data engineers, data scientists, business analysts, and developers as these teams gather data, analyze the data, create ML models, and create real-time analytical solutions. The collaborative platform helps break down common silos resulting in faster insights and innovation in the connected apps and services that are created.
- The AI Toolkit for Azure IoT Edge allows ML and AI models to be created in the cloud where computational power is easily accessible. The resulting ML and AI models can then be delivered directly to Azure IoT Edge solutions allowing real-time event processing and AI-enabled apps to be delivered next to the equipment in the factory.
- Azure Cognitive Services are a set of pre-built AI models that can be integrated into any application or service, including on mobile platforms. Azure Cognitive Services includes AI models for custom vision, computer vision, video intelligence, speech recognition, translation, text-to-speech, text analytics, and language understanding.
What makes Microsoft especially valuable is the size and breadth of the cloud services. Microsoft currently operates more cloud regions than any other vendor and has one of the most complete hybrid cloud services. Microsoft refers to these as the Intelligent Cloud and the Intelligent Edge. The vision is that the device or app location should not matter – the cloud and AI capabilities should be available everywhere.
Using Azure Databricks, organizations can more rapidly analyze their data and create ML models. Those models can be connected into the Azure IoT Edge devices creating intelligent solutions that run directly in the factory. These solutions can then be augmented with Azure Cognitive Services delivering intelligence tools, apps, and insights to run and operate the manufacturing operations.
Just implementing IoT solutions in a factory and manufacturing operation will not be enough to become a leader and stay ahead of the competition. Ongoing success necessitates driving greater intelligence into operations, changing processes, reallocating people resources, and creating new revenue streams. A foundation for these transformative actions will come from fully embracing ML and moving towards AI-enabled and AI-driven solutions. Microsoft’s Azure platform provides one of the strongest foundations to create and deliver these solutions. And as my colleague, David McKnight, discussed in a recent blog, there are bottom-line benefits as well.