How advanced machine learning and existing process data can improve productivity, plant performance and equipment reliability.

At Caverion Industria, our goal to ensure the performance and reliability of industrial equipment, assets and processes while improving safety and productivity of human experts. New insights for identifying faults and production deviations well in advance combined with proactive decision making based on data are the enablers for O&M teams to always stay one step ahead.

Avoiding downtime and process instability in industrial maintenance

Why is unplanned downtime still a major challenge?

A single equipment failure or undetected process issue may incur significant production losses and repair costs if it happens suddenly and at the wrong time. To minimize unplanned downtime, across the entire plant, with dynamic adaptation to complex process characteristics and variations we turn to data, AI and advanced machine learning methods. This means applying advanced mathematical methods and computing capacity specifically developed for industrial process domain.

Why have traditional methods not been successful?

The first major challenge comes from the number of assets, equipment and the range of possible faults, failures and malfunctions in typical industrial process plant. Secondly, the dynamic and complex nature of industrial production, different operating modes, changes in loads, raw material variance and product changes, means there are no easy ways to define fixed alarm limits for condition monitoring. Subtle changes in asset performance, slow trends, leaks, clogs and abnormal operational situations might go undetected for some time.

In general, traditional methods of predictive maintenance and CBM have been too narrowly focused or produce too many unnecessary false alarms while still missing some major events.

Holistic approach to production efficiency and reliability improvement

Supporting both sides of the O&M, across all assets and entire process and providing context-specific and relevant insights; right information, at the right format and at the right time. This means understanding the actual asset condition with the current and predicted performance trends across the entire process, production line or plant. The result is that issues can be effectively analyzed, operators can perform adjustments accordingly, and maintenance teams can plan the repair actions and order spare parts in time.

Confident decisions supported and validated by data include e.g. when is the optimal time to replace parts, what assets need special attention and how do we effectively plan the work remotely and make sure experts can focus on what is important.

Insights and valuable information from data

Most industrial production plants are ready for data-driven operations and advanced analytics. While it may have never been utilized at its full potential before, lots of high-quality data does exist in the automation and DCS systems, and we can easily gather more if necessary.

The process data (flows, pressures, temperatures, etc.) is extremely valuable when we apply machine learning to it and with additional IoT sensors (vibration, audio analytics, machine vision) we can cover a wide range of use cases. With secure connectivity and cloud computing power the data analytics methods can be applied anywhere regardless of geographic location.

At Caverion, our data-driven maintenance solutions consist of three main elements:

1. Machine Learning based broad scope anomaly detection on existing process data
2. Smart IoT sensors and analytics
3. Predictive maintenance & advanced condition monitoring applications

What broad scope anomaly detection does is it takes all available process data and uses machine learning to understand the complex behavior of the plant or production line. This approach understands the dynamics of the process (correlations, seasonal changes, operating modes, load changes, interdependencies of various process parts and measurements etc.) and provides early warnings when a combination of signals in a specific context indicates something abnormal is starting to happen.

Real-time analysis of thousands of measurement signals using machine-learning models makes it possible to identify process deviations at a very early stage.

Instead of focusing only on critical equipment or specific failures, broad scope anomaly detection can flag all types of deviations, from repeating faults to novel, never before seen issues.

The result is comprehensive approach for minimizing unplanned downtime, failures and process inefficiencies. With typically second or minute level average values of signals analyzed, automatic warnings are raised when something unusual is picked up by the computation. This is when the human expertise comes into play, in the form of making the final decision.

What AI and computers can’t do as effectively as a human expert is to diagnose and determine the actions needed to remedy a situation. Warnings raised by the algorithms are provided via web-based UI for human experts to analyze. Most commonly maintenance, process engineers and operators are the key users. Each warning is handled by the expert(s) and feedback is given back to the system in the form of warnings rating (TRUE or FALSE). This is important for the accuracy and adaptability of the machine learning models and answers the question: how to ensure continuous improvement and what happens when something in the process changes?

AI learns constantly from data but also from user feedback. In this way AI helps experts learn more about their process and equipment, while teaching the AI to become more accurate over time.

Each time the user rates a warning as FALSE, the machine learning models are immediately updated accordingly, and will not produce a warning from that type of situation in the future. Before going live with data, automatic training of machine-learning models is done using set of clean, fault-free historical data or by running the models live for training a couple of weeks.

With existing historical data, if available, and automatic learning, the service can be implemented quickly and without requiring a massive amount of time or effort. The key is to incorporate AI based analytics and warnings handling as a part of the daily O&M processes and to highlight the realized benefits – downtime and costly breakdowns avoided – across the organization constantly.

To those process areas where existing data is not available, and to enable remote operations and monitoring anywhere we provide the wireless IoT sensors with accompanying analytics. In addition to vibration monitoring, we provide smart sensors that can analyze anything from temperature to sounds and video, depending on the use case need.

The combination of process data and smart IoT analytics are the foundations of predictive maintenance, improving the asset lifecycle and investment planning processes.

Industrial maintenance: a combination of artificial intelligence and experts

Artificial Intelligence and machine learning methods with big data are here to help the experts make better decisions and to focus on the important tasks in a productive way. AI is at its best for tasks requiring high computational capacity and complexity, such as real-time mathematical analysis of thousands of signals. Human experts outperform the AI in reasoning, interpreting information, weighing options and managing the work of corrective actions. The combination of machine learning based automatic analysis of data and the shared expertise and years of knowledge from operators, process engineers and maintenance technicians is the key to high performing teams.

At Caverion, we combine years of O&M experience from different industries with in-depth expertise in applied mathematics, data science and IoT technologies. With in-depth understanding the needs of the end users at site, usability and practical applicability ensures adoption and fit for purpose. Digitalization is a change that requires good leadership and strong support from various levels of the organization, and companies across multiple industries have shown that it can bring significant benefits. The ROI and time-to-value of applying a proven service with expert support without having to make changes to the existing IT and OT landscape can provide a way to justify further investments.

Data driven production and maintenance can provide significant benefits in productivity and competitive advantage. Whether you are just getting started on the digitalization journey or have already advanced with this transformation, we are happy to help and setup a discussion about this exciting topic.

The original Finnish version of this blog can be found here.