From reactive to proactive: Use anomalies in your operations to feed AI- and ML-based analytics

Posted: June 3, 2024

Industrial organizations today capture a wealth of data from sensors, equipment and production processes. This data resides in disparate systems, often siloed and underutilized. Traditional methods of data analysis struggle to keep pace with the sheer volume and complexity of this information. As a result, valuable insights remain buried, hindering the ability to optimize operations, prevent failures and make data-driven decisions. 

As a long-time supplier of industrial data platforms, AVEVA has seen firsthand how companies are using their data to drive sustainable and profitable changes in their operations

Beyond routine data: unlocking the value of anomalies

Industrial data offers a vast repository of valuable information. The majority of this information comes from data streams generated by various operating assets and systems. While this data provides a baseline for operations control and monitoring, the most valuable insights often lie within the anomalies—the deviations from the norm that signal impending issues or areas for improvement.

Imagine a critical piece of equipment failing unexpectedly, causing a production shutdown. What if subtle changes—such as anomalies in vibration or temperature data—existed beforehand and could signal that the equipment needs maintenance?

Fingrid Oyj, Finland’s central power grid operator, used this kind of anomaly detection to optimize its power system performance. Recognizing the limitations of manual analysis for their massive data sets (over 4 million daily measurement samples), Fingrid Oyj implemented AVEVA™ PI System™, which empowered it to not only automate fault processing, but also delve deeper into anomaly detection. By integrating live data streams with custom algorithms, Fingrid Oyj can now proactively identify disturbances and anomalies in the power grid, enabling them to safeguard critical infrastructure and ensure a more resilient power supply.

Using anomalies in advanced analytics

Artificial intelligence (AI) and machine learning (ML) models can be trained to identify deviations in real time, allowing for proactive intervention. For instance, an AI system could detect a slight increase in vibration data from a specific machine, potentially signaling a developing imbalance. This early warning can empower you to take preventive maintenance actions, potentially avoiding a catastrophic equipment failure later.

The true power of anomalies lies in the insights they unlock with the help of advanced analytics. By feeding historical and real-time data from sensor readings, vibration patterns and other operational metrics into AI and ML models, industrial organizations can identify patterns and predict future issues. This allows organizations to move from reactive maintenance to a predictive approach, optimizing resource allocation and minimizing downtime. Industrial data platforms become even more powerful when coupled with AI and ML, transforming raw data into actionable insights that empower organizations to optimize operations and accelerate industrial intelligence.

Anomaly-driven intelligence

Anomaly records are a form of metadata—“data about data”—and are captured and stored in AVEVA PI System. While anomaly detection is a powerful tool in itself, its value is significantly enhanced with the human expertise of your connected workforce. Human expertise becomes data in the anomaly records, which can then be used by AI/ML at the right time to find the root cause.

Anomaly capture has three important roles:

1. Enriching AI/ML models: By analyzing historical trends, equipment sensor readings and process parameters, the AVEVA PI System can automatically identify potential contributing factors to anomalies. This rich context empowers AI/ML models to learn from a more comprehensive data set, leading to more accurate anomaly identification and future predictions.
2. Building a historical baseline for normal operations: Industrial processes are complex systems with numerous variables. Defining “normal” can be challenging. However, by capturing anomalies over time, the AVEVA PI System helps establish a historical baseline for what constitutes typical equipment behavior. This baseline serves as a crucial reference point for AI/ML algorithms. Deviations from this baseline can then be flagged as potential anomalies, allowing the AI/ML model to focus its analysis on these critical events.
3. Facilitating root cause analysis: Simply identifying anomalies is only half the battle. Understanding the root cause behind them is essential for taking corrective action. The AVEVA PI System’s anomaly capture capabilities extend beyond just the anomaly itself. It can capture surrounding data points like timestamps, or sensor readings from nearby equipment and process parameters. This rich contextual data empowers AI/ML models to not only detect anomalies but also to analyze the contributing factors that led to them. This deeper understanding allows for more effective troubleshooting and preventative maintenance strategies.

However, it is crucial to remember the adage “garbage in, garbage out.” The accuracy of your AI and ML results hinges on the quality of your data.

AVEVA PI System products facilitate the capture and storage of anomalies and reliable contextual information and delivers it to AI & ML models. As these models encounter a greater volume of anomalies over time, they develop a heightened capacity to identify future occurrences, fostering a continuous learning cycle. With the crucial information provided by your AVEVA PI System, you can differentiate your AI and ML results from those of your competitors and consistently outperform in your market.

The powerhouse combination of reliable historical data and analytics

As industrial data management continues to evolve, the ability to extract value from anomalies is a true differentiator. While powerful analytics engines and AI technology can help you uncover insights not readily seen in daily operations, large volumes of reliable, highly granular data is required as input. Clear and comprehensive data is essential to uncovering hidden patterns and identifying root causes.

An industry leading data platform like AVEVA PI System combines asset health, process performance, and the connected worker experience into organized datasets. Then, with the help of AI- and ML-based analytics, organizations like Fingrid Oyj can identify degradations before they cause damage, and then take proactive measures. By leveraging the combination of reliable historical data and advanced analytics, organizations can optimize resource allocation, maximize equipment lifespan and, ultimately, achieve operational excellence.