Building Explainable Predictive Maintenance with Machine Learning and Expert Systems

Unexpected equipment failures remain one of the most significant and costly challenges facing manufacturers today. A single unplanned outage can result in lost production, missed delivery schedules, increased maintenance costs, reduced product quality, and even safety risks. For many organizations, the financial impact extends far beyond the cost of replacing a failed component; it can disrupt entire production schedules and affect customer satisfaction.

Historically, manufacturers have relied on reactive (run-to-failure) or preventive maintenance strategies. Reactive maintenance delays until equipment fails, often resulting in costly downtime and emergency repairs. Preventive maintenance improves reliability by servicing equipment on a fixed schedule, but it can also lead to unnecessary maintenance, premature component replacement, and missed failures that develop between inspections.

The global predictive maintenance market

https://www.grandviewresearch.com/industry-analysis/predictive-maintenance-market

As a result, manufacturers are increasingly adopting predictive maintenance, a data-driven approach that monitors equipment health and identifies developing problems before they lead to failure. This shift is accelerating rapidly across the industrial sector. According to market research, the global predictive maintenance market was valued at $13.65 billion in 2025 and is projected to grow to $97.37 billion by 2034, reflecting the increasing demand for technologies that improve asset reliability, reduce downtime, and optimize maintenance resources.

Predictive maintenance is powered by predictive analytics, which combines statistics, artificial intelligence (AI), data analytics, and machine learning to analyze historical and real-time operational data and forecast future equipment behavior. By recognizing patterns that precede failures, predictive analytics enable maintenance teams to move from reacting to problems to proactively preventing them.

However, successful predictive maintenance requires more than simply applying machine learning algorithms. Industrial environments are complex, and effective decisions must also incorporate engineering expertise, equipment knowledge, and operational context. The most successful solutions combine data-driven predictions with the practical knowledge that maintenance and operations teams have developed over years of experience.

How can manufacturers combine machine learning with engineering expertise to create a predictive maintenance solution that is both accurate and explainable? Before exploring how ADISRA SmartView addresses this challenge, let us first look at what predictive maintenance really is and why it has become such an important strategy for modern manufacturing.

What Is Predictive Maintenance?

Predictive maintenance is a maintenance strategy that uses historical and real-time operational data to assess equipment health and predict when maintenance should be performed. Rather than servicing assets on a fixed schedule or waiting for them to fail, predictive maintenance allows maintenance activities to be scheduled based on the equipment's actual condition.

This condition-based approach helps organizations reduce unplanned downtime, extend equipment life, optimize maintenance resources, and lower operating costs while improving overall equipment reliability.

However, one of the greatest challenges in predictive maintenance is transforming enormous amounts of operational data into meaningful, actionable information. Industrial facilities have long been data-rich but information-poor. Collecting data is no longer the problem; understanding what the data means and knowing how to respond is where the real value lies.

This challenge is what led us to the design philosophy behind ADISRA SmartView. We believe the most valuable place to generate predictive intelligence is at the edge, close to the machine where the data is created. Processing data at the source enables faster decisions, reduces unnecessary data movement, and allows operators to respond before minor issues become major failures.

To support this approach, ADISRA SmartView integrates both a Machine Learning Module and a Rule-Based Expert System directly into its HMI/SCADA development environment. Together, these technologies transform raw operational data into explainable, actionable maintenance recommendations.

Choosing the Right Machine Learning Algorithm

One of the most common misconceptions about predictive maintenance is that there is a single machine learning algorithm that can predict equipment failures. In reality, predictive maintenance is an application that combines multiple machine learning techniques, each designed to answer a specific question about an asset's health.

Some algorithms forecast how equipment will behave in the future, while others determine whether a failure is likely to occur, identify the type of failure, or detect subtle changes in operating conditions that may indicate a developing problem. Selecting the appropriate algorithm depends on the equipment being monitored, the data available, and the business question you are trying to answer.

ADISRA SmartView includes several built-in machine learning algorithms that can be applied individually or together to create predictive maintenance solutions. The table below summarizes how each algorithm contributes to monitoring equipment health.

Although each algorithm has its own strengths, they are often most effective when used together. A forecasting model might predict that a motor's temperature will exceed its normal operating range within the next 24 hours, while an anomaly detection model confirms that the trend is unusual compared to historical operation. A rule-based expert system can then evaluate these results alongside engineering knowledge, such as motor load or operating limits, to determine whether maintenance should be scheduled.

In the following example, we will demonstrate this hybrid approach by combining Forecasting, Time-Series Anomaly Detection, and ADISRA SmartView's Rule-Based Expert System to identify a developing motor problem before it results in an unexpected failure.

Predictive Maintenance in Action: Monitoring an Industrial Electric Motor

To illustrate how this hybrid approach works in practice, let us examine one of the most common and critical assets found in virtually every manufacturing facility: the industrial electric motor.

Electric motors are electromechanical devices that convert electrical energy into mechanical motion, providing the power required to operate countless industrial processes. They drive pumps, compressors, conveyors, fans, blowers, mixers, machine tools, and automated production equipment, making them the workhorses of modern manufacturing. Their widespread use and importance to plant operations also make them strong candidates for predictive maintenance.

A single motor failure can significantly impact production. Depending on the application, an unexpected failure may stop an entire production line, interrupt a critical process, delay customer shipments, and result in costly emergency repairs. For this reason, continuously monitoring motor health and identifying developing problems before they lead to failure can provide substantial operational value.

A hybrid motor-monitoring strategy combines forecasting, time-series anomaly detection, and a rule-based expert system. Each technology plays a different role in the predictive maintenance process, and together they provide a more complete and explainable approach than any one technique used alone.

Forecasting analyzes historical sensor data to predict future equipment behavior. For an electric motor, a forecasting model can estimate future temperature, vibration, current draw, or energy consumption. This allows maintenance teams to understand where motor health is heading over the next several hours or days, rather than relying only on current measurements.

Time-Series Anomaly Detection evaluates operating data over time to identify unusual patterns, trends, or deviations from expected behavior. Instead of depending only on fixed alarm limits, it can detect subtle changes such as a gradual increase in vibration, an unexpected rise in temperature, or a shift in current draw that may indicate the early stages of equipment degradation.

The Rule-Based Expert System applies engineering knowledge through configurable if-then rules. It evaluates the machine learning results within the context of the motor’s actual operating conditions. By considering factors such as motor load, operating limits, alarm history, and maintenance experience, the expert system can validate predictions, reduce false alarms, identify likely failure conditions, and recommend appropriate maintenance actions.

Together, these technologies transform raw sensor data into actionable intelligence:

Forecasting predicts what is likely to happen.

Anomaly detection determines whether the behavior is unusual.

The Rule-Based Expert System evaluates the operating context and recommends what should happen next.

Implementing the Hybrid Predictive Maintenance Solution

Implementing this hybrid approach in ADISRA SmartView begins by creating a Machine Learning Document and selecting the historical operating data to train the machine learning models. The quality of this historical data is critical because it forms the knowledge base from which the models learn normal operating patterns, long-term trends, and the early indicators of equipment degradation.

Building the predictive maintenance solution is a straightforward process:

Select the training data. Choose the input variables that describe the motor's operating condition, such as temperature, vibration, current draw, motor load, speed, and runtime, and define the target value the model should predict.

Train a Forecasting model. The forecasting model analyzes historical operating data to learn how key measurements change over time. Once trained, it can estimate future values such as motor temperature or vibration, providing early warning of conditions that may lead to failure.

Train a Time-Series Anomaly Detection model. While forecasting predicts where the motor's condition is heading, the anomaly detection model continuously evaluates incoming data to determine whether the current operating behavior deviates from normal historical patterns.

Deploy trained models. The models are added directly to the ADISRA SmartView application, where they are linked to live runtime tags. Predictions can be executed on demand, at scheduled intervals, or automatically as new sensor data becomes available.

Together, these models provide both a prediction of future equipment behavior and an assessment of whether the observed behavior is abnormal. Their outputs become valuable inputs to the next stage of the hybrid solution, the Rule-Based Expert System.

Validating Machine Learning Predictions with the Rule-Based Expert System

Machine learning is extremely effective at recognizing patterns and identifying potential problems, but not every prediction should immediately trigger a maintenance action. Industrial equipment operates under constantly changing conditions, and engineering judgment is often required before determining that a developing trend truly represents a fault.

This is where the Rule-Based Expert System adds another layer of intelligence.

The expert system evaluates the machine learning results alongside real-time operating conditions. It has access to model outputs such as prediction score, confidence, probability, affinity, anomaly indicators, and other machine learning metrics. These values are combined with live process information, such as motor speed, load, operating limits, and temperature, to determine whether the prediction is reliable enough to warrant maintenance.

For example, an anomaly detection model may identify abnormal vibrations while the motor is operating. Rather than immediately generating an alarm, the Rule-Based Expert System evaluates additional operating conditions using engineering logic.

Example Rule: Predictive Maintenance for an Industrial Electric Motor

IF

Motor temperature is greater than 90°C

AND vibration exceeds 4.5 mm/s RMS

AND current imbalance between phases is greater than 5%

THEN

Conclude that the motor is experiencing a high risk of failure, with stator winding insulation degradation as a likely cause.

Recommended Actions

Reduce the motor load if possible.

Schedule maintenance at the next available opportunity.

Perform insulation resistance and winding resistance tests to confirm the diagnosis.

Conversely, if the prediction confidence is low or the required operating conditions are not satisfied, the expert system can suppress the alarm, continue monitoring the equipment, or request additional evidence before recommending maintenance.

This hybrid approach combines the strengths of both technologies:

Machine Learning discovers patterns, forecasts future behavior, and detects anomalies that may not be visible to an operator.

The Rule-Based Expert System applies engineering knowledge to validate those predictions, reduce false alarms, and determine the appropriate maintenance response.

In other words, machine learning provides insight, while the Rule-Based Expert System provides the confidence to act. Together, they create an explainable, trustworthy predictive maintenance solution that helps manufacturers reduce downtime, improve maintenance decisions, and increase asset reliability.

Conclusion

Predictive maintenance is about much more than predicting when a machine might fail. Its real value lies in transforming operational data into actionable decisions that improve equipment reliability, reduce maintenance costs, and minimize unplanned downtime.

Machine learning has become an essential tool for identifying patterns, forecasting equipment behavior, and detecting anomalies that may indicate developing failures. However, predictions alone are not enough. Maintenance teams need to understand why a prediction was made and what action should be taken.

That is where ADISRA SmartView's hybrid approach delivers a unique advantage. By combining Forecasting, Time-Series Anomaly Detection, and an integrated Rule-Based Expert System, ADISRA SmartView produces predictions that are both accurate and explainable. Machine learning uncovers hidden patterns within operational data, while engineering rules validate those predictions, reduce false alarms, and translate analytical results into practical maintenance recommendations.

As manufacturers continue to embrace Industry 4.0 and Industrial AI, the organizations that gain the greatest value will not simply collect more data; they will transform that data into reliable, explainable intelligence that supports better operational decisions.

With ADISRA SmartView, predictive maintenance becomes more than an AI model. It becomes a practical decision-support system that combines data science with engineering expertise to help manufacturers improve asset reliability, increase productivity, and prevent unexpected equipment failures.

Ready to Experience ADISRA SmartView?

Download your free evaluation copy of ADISRA SmartView here.

Would you like a personalized demonstration? Simply click the link to request a private demo, and a member of our team will contact you to schedule a one-on-one demonstration tailored to your application.

ADISRA®, ADISRA's logo, InsightView®, KnowledgeView® and RapidView™ are registered trademarks and trademarks of ADISRA, LLC.