Building a Closed-Loop Reliability Program with Maximo Health, Predict, and Reliability Strategies

The Maximo Application Suite promise for asset performance management is a closed loop: failure analysis drives reliability strategy, strategy drives maintenance work, work history drives health scores, health scores feed predictive models, and predictive models inform the next round of investment decisions. In practice, most organizations operate a fragmented loop. FMEAs live in spreadsheets, PMs are based on calendar intervals, health scores are computed monthly in a data warehouse, and IoT sensor data is siloed in a separate monitoring platform. MAS 9.1 changes that by making Health, Predict, Monitor, Reliability Strategies, and Manage part of a single operational capability, not separate products.

This article walks through the architecture of a closed-loop reliability program, the configuration of Health and Predict, the use of Reliability Strategies to operationalize FMEA, and the patterns that turn a deployment from a dashboarding exercise into a decision-making system that measurably reduces unplanned downtime.

The Reliability Stack: A Capability View

MAS 9.1 treats reliability as a stack of capabilities rather than a collection of products. At the foundation is Maximo Manage, which holds the asset master, work history, failure codes, job plans, and PMs. Above that is Maximo Monitor, which ingests IoT sensor data from PLCs, SCADA systems, and field devices. Monitor processes that data through analytics pipelines and produces signals, anomalies, and threshold breaches. Above Monitor is Maximo Predict, which uses AI and machine learning to forecast failure probability, remaining useful life, and degradation patterns. Above Predict is Maximo Health, which fuses the data from all sources into composite health scores, criticality rankings, and risk profiles. The application layer is Reliability Strategies, which connects FMEA, RCM, and the curated reliability library to the work execution in Manage.

The value of this stack is the data flow. Manage provides asset master and failure history. Monitor provides real-time sensor data. Predict provides forward-looking estimates. Health provides prioritization. Reliability Strategies provides the bridge from prediction to action. Without the integration, each layer produces its own dashboard. With the integration, the layers produce a decision.

The "AI Service 9.2" update introduced in January 2026 swaps the underlying LLM for predictive condition classification, maintenance component classification, and FMEA reasoning. Health features that consume AI Service, including asset Q&A, recommended actions, and FMEA-driven indicator generation, inherit the upgrade automatically. No Health model reconfiguration is required. This is the kind of platform improvement that pays back without effort from the customer.

Health Scoring: The Anatomy of a Model

Maximo Health computes health as a composite of multiple component scores. A typical asset health model might include four or five components: Condition, Reliability, Performance, Environment, and (optionally) Predictive. Each component is itself a weighted average of indicators. Each indicator is computed from one or more data sources.

The Condition component typically draws from inspection results, meter readings, and condition monitoring data. A motor might have indicators for vibration level, temperature, insulation resistance, and oil analysis. Each indicator is scored against a defined range: green (good), yellow (warning), orange (degraded), red (failed). The component score is a weighted average of the indicator scores.

The Reliability component draws from failure history. A high failure rate over the past 12 months lowers the Reliability score. The component uses the failure data from Manage to compute a time-since-last-failure and a frequency-of-failure metric. These feed into the composite.

The Performance component draws from operating data. An asset that operates outside its design envelope (too hot, too cold, too high a load) accumulates Performance penalties. The data typically comes from meters and from the operating context tags in Monitor.

The Environment component captures ambient factors. For an outdoor asset, weather and seasonal effects matter. For an indoor asset, dust, humidity, and temperature matter. The Environment component is the most asset-class-specific, and many deployments choose to omit it for assets where ambient factors are not material.

The Predictive component, when Monitor and Predict are integrated, draws from the failure probability and remaining useful life estimates. A high predicted failure probability lowers the score. A short remaining useful life lowers the score. The Predictive component transforms Health from a backward-looking score into a forward-looking one.

A sample Health model configuration might look like this:

apiVersion: health.maximo.ibm.com/v1
kind: HealthModel
metadata:
  name: centrifugal-pump-health
  namespace: mas-health
spec:
  assetClass: PUMP-CENTRIFUGAL
  components:
    - name: Condition
      weight: 0.30
      indicators:
        - name: Vibration
          source: monitor.signal.vibration.rms
          ranges:
            green: "<2.8 mm/s"
            yellow: "2.8-4.5 mm/s"
            orange: "4.5-7.1 mm/s"
            red: ">7.1 mm/s"
        - name: BearingTemp
          source: monitor.signal.temp.bearing
          ranges:
            green: "<60C"
            yellow: "60-70C"
            orange: "70-80C"
            red: ">80C"
    - name: Reliability
      weight: 0.20
      source: manage.failure.history
      window: 365
    - name: Performance
      weight: 0.20
      source: monitor.signal.flow.deviation
    - name: Predictive
      weight: 0.30
      source: predict.failure.probability
      window: 30
  refreshInterval: 15m
  alertThreshold: orange

This model assigns 30 percent weight to Condition, 20 percent to Reliability, 20 percent to Performance, and 30 percent to Predictive. The model refreshes every 15 minutes and triggers an alert when any indicator drops to orange. The specific weights and ranges should be tuned to your asset class, your operating context, and the historical data that informs your reliability practice.

Integrating Monitor and Predict: The Heartbeat Model

Health with Manage data alone is valuable. Health with Monitor and Predict integrated is transformative. The difference is the difference between a snapshot and a heartbeat. A snapshot tells you what the asset looked like when the inspection was performed, a month ago. A heartbeat tells you what the asset looks like right now and what it will look like tomorrow.

The integration has three layers. The first layer is the data flow from Monitor to Health. Monitor processes high-frequency IoT data and produces signals. A signal is a derived metric such as a rolling average, a rate of change, a statistical deviation, or a threshold breach. These signals map directly to Health indicators. The "Use Condition Monitoring" checkbox on a meter-based scoring factor aligns the indicator thresholds with the IoT sensor ranges and enables real-time updates. With this checkbox enabled, Health indicators update when new sensor readings arrive, not just on manual inspection cycles.

The second layer is the data flow from Predict to Health. Predict uses historical and real-time data to compute failure probability within a defined window, remaining useful life, and anomaly scores. These outputs map into Health indicators in the same way Monitor signals do. The output from Predict, for example a "30-day failure probability of 18 percent," can be a Health indicator on its own or can feed into the Predictive component.

The third layer is the closed-loop action. When a Health score drops to a threshold, the system can automatically create a Health Work Queue in the Work Queue Manager. A planner picks up the work queue, reviews the asset's health profile, and creates a work order. The work order references the failure codes that drove the score down. The work order executes. The asset's health recovers. The model learns.

A common pattern is to set the alert threshold at orange. When an asset hits orange, an automated work queue is populated. When an asset hits red, an automated notification is sent to the reliability engineer. The thresholds are tunable. The principle is that the threshold should trigger action, not just awareness. If an orange alert never leads to a work order, the threshold is wrong.

Reliability Strategies: Operationalizing FMEA

FMEA is a 70-year-old discipline. The challenge has never been the analysis itself; it has been the connection between the analysis and the work. Most organizations produce beautiful FMEA documents that nobody reads after the workshop. Maximo Reliability Strategies solves this by writing the FMEA directly into the work objects that technicians execute.

The Reliability Strategies library contains over 58,000 failure modes and corresponding mitigation activities, developed by industry and domain experts. The library covers centrifugal pumps, motors, valves, compressors, conveyors, transformers, and hundreds of other asset types. Each entry includes the failure mode, the failure mechanism, the failure influence, the operational context, the recommended mitigation activity, the recommended frequency, the labor hours, and the effectiveness rating.

The workflow is straightforward. A reliability engineer selects an asset, selects the asset type, and selects the asset configuration. The library returns the failure modes that apply. The engineer reviews the failure modes, accepts or rejects each one, and the system generates the corresponding PMs and job plans. The PMs and job plans are written to Manage, where they become the maintenance schedule.

A typical Reliability Strategies session might involve a 1,000-horsepower centrifugal pump. The library returns failure modes for the mechanical seal, the bearings, the coupling, the impeller, the casing, and the lubrication system. Each failure mode has a recommended PM: vibration analysis quarterly, oil analysis monthly, bearing temperature monitoring continuous, seal flush flow verification weekly. The reliability engineer accepts the recommendations and the system generates 12 PMs and corresponding job plans. The work that would have taken weeks of analysis is completed in hours.

The MAS 9.1 enhancement added AI to the workflow. The AI can suggest boundary conditions for the analysis (operating limits, environmental conditions, duty cycle). The AI can generate components and failure mechanisms. The AI can propose additional failure modes that the engineer might not have considered. The AI assistance is most valuable for organizations that are new to FMEA and do not yet have deep in-house reliability expertise.

Asset Investment Planning: The Strategic Layer

Reliability Strategies produces the tactical maintenance plan. Asset Investment Planning (AIP) produces the strategic investment plan. AIP, which is new in MAS 9.1, enables users to generate and compare multiple investment scenarios based on business priorities, with rapid weighted analysis and multi-objective optimization to balance cost, risk, and performance.

AIP operates on a multi-year horizon. The planner defines a portfolio of assets, a set of candidate interventions (replacement, refurbishment, run-to-failure, change of maintenance strategy), and a set of constraints (budget, downtime, risk tolerance). AIP computes the optimal intervention sequence for each asset, given the constraints, and rolls the results up to a portfolio view. The planner can compare scenarios side by side and choose the plan that best fits the organization's strategy.

The data inputs to AIP are drawn from Health, Predict, Reliability Strategies, and Manage. The asset condition and risk scores come from Health. The failure probability and remaining useful life come from Predict. The maintenance strategy recommendations come from Reliability Strategies. The cost data and capital planning constraints come from Manage. AIP synthesizes these inputs into a defensible investment plan that traces every recommendation back to the underlying reliability data.

For executives, AIP answers the question: "If we have $10 million to spend on asset interventions over the next three years, which assets should we replace, which should we refurbish, and which should we keep running?" The answer is data-driven, not politics-driven.

Common Pitfalls and Field-Tested Patterns

The first pitfall is over-instrumenting. Not every asset needs 10 sensors. The cost of instrumentation, the cost of data storage, and the cost of model maintenance all scale with sensor count. Focus IoT investment on high-criticality assets where the data changes decisions. A 5,000-horsepower gas turbine deserves a full sensor suite. A 5-horsepower utility pump does not. A useful rule of thumb is to instrument the top 10 to 20 percent of your asset portfolio by criticality, and to use the data from those assets to inform the maintenance strategy for the long tail.

The second pitfall is trusting stuck sensors. A vibration sensor that reports "OK" for three months because it has failed creates false confidence. The health score for an asset with a stuck sensor will look better than it should. Implement sensor health monitoring. Validate sensor readings against expected ranges. Flag sensors that have not changed value in a defined window. Replace failed sensors quickly.

The third pitfall is skipping the validation loop. Predictive models degrade over time as operating conditions change. A model trained on data from a normal operating year will not perform well during a year with abnormal ambient conditions, a process change, or a different product mix. Establish a retraining cadence. Monitor the model's accuracy. When accuracy drops below a threshold, retrain. The Maximo AI Service supports automated retraining, but the cadence and the threshold are your decisions.

A field-tested pattern is the "one new indicator" rule. When adding predictive indicators to a Health model, introduce them one at a time. Validate each indicator's impact on the health score before making it permanent. Communicate changes to all stakeholders. Maintain a changelog. This discipline prevents the "dashboard rot" that comes from adding too many indicators at once and losing the ability to understand the score.

Another field-tested pattern is to define ownership boundaries clearly. Who owns the sensor? Who owns the model? Who owns the indicator? Who owns the threshold? Who owns the action? When ownership is ambiguous, the system degrades. A sensor goes uncalibrated. A model goes unretrained. A threshold goes untuned. An action goes unexecuted. Assign owners for each layer and hold them accountable.

Practical Implications

For reliability engineers, the closed-loop reliability program means new tools and new responsibilities. The FMEA workshop is now a structured process in Reliability Strategies, with AI assistance, library content, and direct output to PMs. The reliability engineer spends less time authoring spreadsheets and more time validating the analysis. The reliability engineer's KPIs shift from "number of FMEAs completed" to "percentage of critical assets with up-to-date strategies."

For asset managers, the closed-loop program means data-driven prioritization. The asset review meeting is no longer driven by instinct or by the loudest stakeholder. It is driven by the Health score, the Predict forecast, the failure history, and the investment analysis. The asset manager can present a defensible plan that traces every recommendation back to the underlying data.

For IT teams, the closed-loop program means integrating data flows between Manage, Monitor, Predict, and Health. The integration is configured in MAS, but the operational monitoring is the same as any other data pipeline. Monitor the data flow. Alert on broken integrations. Validate the data quality. The reliability program is only as good as the data that feeds it.

For finance teams, the Asset Investment Planning module changes the conversation about capital allocation. Instead of negotiating from a position of "we need $20 million," the conversation is "here are three scenarios with different risk profiles, here is the optimal plan under each, and here is the data that supports each." The conversation is more efficient, more transparent, and more defensible.

Bottom Line

The closed-loop reliability program in MAS 9.1 is achievable, but it is not automatic. It requires careful configuration of Health models, deliberate integration of Monitor and Predict, disciplined use of Reliability Strategies to operationalize FMEA, and clear ownership of every layer from sensor to action. The payoff is significant. Organizations that close the loop see a measurable reduction in unplanned downtime, an extension of asset life, and a defensible capital allocation process.

The path forward is to start with Health and Manage data alone. Build a Health model for your most critical asset class. Score your assets. Identify the ones that need attention. Create work orders. Validate the impact. Once the Health model is producing actionable results, integrate Monitor. Add real-time condition data. Once the Monitor integration is mature, add Predict. Add forward-looking failure probability. Once Predict is calibrated, deploy Reliability Strategies. Generate PMs from the library. Connect the PMs to the work execution. Once Reliability Strategies is operational, add Asset Investment Planning. Build the strategic layer. Close the loop.

Each step is a project. Each project delivers value. The closed-loop program is a journey, not a destination. Start now.