Predictive Analytics Pipelines: Real-World AI, Predictive Maintenance, and Time Series Data

By Jason Myers / Developer
Jul 23, 2024

Real-world AI

There’s so much talk about AI these days that it seems we quickly forget that AI isn’t a single type of technology. It’s a category, almost an umbrella term for a wide range of different technologies, applications, and approaches. The terms “Generative AI” and “Machine Learning AI” (often referred to as “Real-World AI”) describe two different branches that fall under the broader AI heading.

Generative AI refers to models and algorithms designed to generate new content, such as text, images, music, or even code similar to the data used to train those models. However, we’re more concerned with real-world AI. This machine learning uses algorithms and statistical models that rely on patterns and inference to enable computers to perform specific tasks without explicit instructions. When we talk about creating smarter systems, real-world AI is the driving force behind them.

Industrial operations constantly strive for optimization and improvement. Even a 1% increase in efficiency can translate into millions of dollars in savings for a company. Real-world AI allows companies to use data to drive improvements and boost their bottom line. In the industrial sector, data-based predictions can have a huge impact. Let’s take a look at what goes into prediction processes and then consider a real-world example.

Predictive analytics

Predictive analytics lets organizations ‌foresee future events and outcomes. This allows businesses to enhance operational efficiency, mitigate risks, and secure a competitive advantage.

Understanding Predictive Analytics

Predictive analytics utilizes statistical models and data analysis to predict future events based on historical and current (in the case of time series data, real-time) data. This approach helps identify potential risks and opportunities, enabling proactive decision-making. Companies don’t need to rely on traditional methods based on ‘anec-data’ of past experiences or intuition.

Predictive analytics help companies across a wide range of verticals to streamline operations, reduce downtime, and allocate resources efficiently. For example, in manufacturing, predictive analytics can forecast equipment failures through the analysis of maintenance records and sensor data, allowing for timely maintenance and reduced operational disruptions.

Key Benefits of Predictive Analytics in Industrial Operations

When implemented and used correctly, predictive analytics provide significant advantages to organizations in the industrial and manufacturing sectors.

• Reduced Downtime: Reactive maintenance, which is basically waiting for something to break, is extremely disruptive. By predicting equipment failures, organizations can schedule maintenance proactively. This ensures that they can take steps to minimize the impact of taking a production line offline, better coordinate service personnel, and enhance productivity overall.
• Increased Operational Efficiency: Overall equipment effectiveness (OEE) is a key metric for industrial operators. The more efficient and consistent production machinery is, the better it is for companies. Predictive analytics identifies inefficiencies and bottlenecks in complex industrial processes. This allows operators to pinpoint trouble areas and make more informed decisions.
• Improved Product Quality: Related to the previous point, the more consistent and reliable industrial equipment is, the more consistent their output will be. Real-time data analysis helps identify production anomalies, ensuring high-quality outputs. This reduces waste, and saves money.
• Enhanced Inventory and Supply Chain Management: The ripple effect of predictive analytics impacts the supply chain as well. Companies can better understand and forecast the demand for raw materials. Optimizing inventory purchasing, storage, and management can reduce costs and improve service levels.

Predictive analytics in action

Having covered some of the key benefits of predictive analytics, let’s now quickly survey some of the techniques and approaches that companies use to achieve them.

As we move deeper into Industry 4.0, using AI becomes increasingly common. Predictive analytics in an industrial context involves using historical data, statistical algorithms, and machine learning techniques to predict future outcomes.

Industrial sensors and systems generate all kinds of data, and most of it is time series data (i.e., data with a timestamp). Tracking change over time is critical to establish baselines and understand how machinery functions compared to those baselines.

Time series data and machine learning are the foundation of predictive analytics. Let’s take a look at a real-world example of predictive analytics in the form of predictive maintenance. InfluxDB is a purpose-built time series database used to manage and process time series data.

LBBC Technologies

LBBC Technologies is the world’s leading designers and manufacturers of industrial autoclave technology. Aerospace customers use this equipment in the manufacture of high-performance castings, like turbine blades. LBBC provides support for customers all over the world. All LBBC equipment comes fitted with industrial gateways that simplify data connections between industrial PLCs and web services, like AWS. LBBC uses these technologies to offer ‘Connected Support’ and Web SCADA to its customers.

We can see real-world AI and time series data at work on the Core Leaching machines LBBC supplies to many aerospace manufacturers. The core leaching process uses potassium hydroxide, a dangerous and costly chemical. During the process, the potassium hydroxide slowly generates silicates that impair the machine’s ability to leach cores as they gradually build up.

High silica levels increase the likelihood of quality defects, but measuring silica levels and predicting when a customer will need to replace the potassium hydroxide is a technical challenge. Using InfluxDB to collect and visualize data from these core leaching machines, LBBC spotted a data pattern that allowed them to monitor potassium hydroxide conditions without complex, sensitive, and very expensive online analysis equipment or cumbersome laboratory tests.

Ultimately, what the data showed was that the build-up of silicates impacted the naturally low vapor pressure of strong hydroxides. Having identified this relationship, LBBC uses InfluxDB to process data to quantify hydroxide quality instead of making mechanical changes to the machinery itself, or investing in other detection methods.

Over the course of a year, LBBC collected data to use for their calculations, including process details about pressure, temperature, valves, pumps, and other important characteristics. The data points spanned 150 cycles with both fresh and spent potassium hydroxide and constituted over 1.3 million data points. Figure 1 shows ‌a data visualization for one of these cycles. The data processing algorithm isolates certain sets of data for each cycle and applies least-squares regression to generate a new variable called Resting Vapor Pressure (RVP).

Figure 1

Using this algorithm, LBBC confirmed that RVP increases as silicates build up and then falls when the customer refreshes the potassium hydroxide supply. Armed with the ability to generate this information from time series data, LBBC tracks RVP and proactively notifies customers when RVP reaches a level that indicates that changing potassium hydroxide will mitigate quality issues.

Figure 2

LBBC Technologies continues to refine their predictive maintenance approaches and best practices as they gain deeper expertise working with time series data.

To get started working with your time series data, try InfluxDB today.