Maximo APM in 9.2: From Condition Monitoring to AI-Driven Reliability

IBM's Maximo Asset Performance Management stack in 2026 is the closest the industry has come to closing the gap between condition monitoring and maintenance execution. The MAS 9.x architecture integrates APM capabilities (Monitor, Health, Predict, Visual Inspection) directly with the EAM application (Manage), creating a closed-loop process from condition monitoring to work order execution. The introduction of Maximo Condition Insight in late 2025, and its maturation through MAS 9.2 in June 2026, adds an agentic AI layer that interprets asset data and recommends corrective actions in natural language.

This deep dive covers the full APM stack in MAS 9.2: Monitor for IoT data ingestion, Health for asset condition scoring, Predict for failure forecasting, Condition Insight for AI-driven interpretation, Reliability Strategies for FMEA and RCM, and Visual Inspection for computer vision. For each component, we will cover what it does, how it is configured, and how it connects to the other parts of the stack. We will also look at the new industry accelerators and the Weibull analysis capability that IBM has added to Manage 9.2.

The APM Stack: Components and Architecture

The Maximo APM stack is a set of integrated applications that work together to move maintenance organizations from reactive to predictive. Each application has a specific role, and the applications are designed to work together as a closed-loop system.

Maximo Monitor is the data ingestion layer. It receives time-series data from IoT sensors, historians, and manual meter readings. Monitor supports 1,600-plus pre-configured device partners, which means that many common industrial sensors can be connected without custom integration. In MAS 9.2, Monitor can run standalone without IoT or Kafka, which simplifies deployment for organizations that do not need the full IoT pipeline. The Edge Data Collector enables offline and edge processing, and new SNMP connectivity expands the range of connectable devices. Hierarchical analytics can roll child metrics up to the parent level, so a parent asset can show aggregated metrics from all its child assets.

Maximo Health is the asset condition scoring layer. It calculates a health score for each asset based on contributors that you define. The default scores available out of the box include: Health (based on open service requests, open corrective work orders, and remaining useful life), Criticality (normalization of priority value), and Risk (calculated as (100 minus health) divided by 100, multiplied by criticality). You can add custom contributors based on meter readings, inspection results, or any other data source that Health can access.

Maximo Predict is the failure prediction layer. It uses AI and machine learning to forecast asset failures before they occur. Predict analyzes historical failure data from Manage and sensor data from Monitor to build predictive models that estimate days to failure, probability of failure, and failure mode. The predictive models are trained on your historical data, which means the predictions are specific to your assets and your operating conditions, not generic industry benchmarks.

Maximo Condition Insight is the AI interpretation layer. It sits above all the other applications and interprets the data they produce. Condition Insight brings together work orders, inspections, meter readings, and reliability strategies to identify patterns in asset behavior and recommend corrective actions. It is the flagship AI capability in MAS 9.x and the component that transforms the APM stack from a set of analytical tools into an intelligent decision support system.

Here is a simplified view of the APM data flow:

Sensors / Historians / Manual Readings
         |
         v
    Maximo Monitor (data ingestion, anomaly detection)
         |
         v
    Maximo Health (condition scoring, health scores)
         |
    +----+----+
    |         |
    v         v
Maximo    Maximo
Predict   Condition Insight
(failure  (AI interpretation,
forecast)  recommendations)
    |         |
    +----+----+
         |
         v
    Maximo Manage (work order creation, execution)
         |
         v
    Maximo Mobile (field execution, inspection, completion)
         |
         v
    Data flows back to Monitor (closing the loop)

The closed-loop architecture is what makes the Maximo APM stack different from standalone condition monitoring or predictive analytics tools. Data flows from sensors through Monitor to Health and Predict, which generate alerts and predictions. Condition Insight interprets the data and recommends actions. Manage creates work orders based on those recommendations. Mobile executes the work in the field. The results of the work (failure codes, actuals, meter readings after repair) flow back into the system, updating the health scores and predictive models. This is the closed loop: monitor, analyze, recommend, execute, measure, repeat.

Maximo Condition Insight: The Agentic AI Layer

Maximo Condition Insight is the most significant AI capability in MAS 9.x. Introduced in late 2025 and matured through 9.2, it is an agentic AI capability within Maximo APM that interprets asset data to explain asset condition, highlight emerging trends, and recommend corrective actions. It works in concert with the other MAS applications to deliver a unified, condition-based maintenance approach.

The engine is powered by IBM watsonx, and the stated aim is to compress the distance between an unusual signal and a defensible maintenance decision. Condition Insight has four headline capabilities:

Summarizing asset condition. Condition Insight reads meter readings, KPIs, and alerts from across the APM stack and produces a natural language summary of the asset's current condition. Instead of a reliability engineer manually reviewing multiple dashboards and reports, the agent does the analysis in seconds and returns a readable explanation.

Mapping conditions to failure modes. The agent maps detected conditions to the failure modes defined in your reliability strategies and FMEA records. This ensures that the recommended action aligns with your organization's maintenance strategy. If the condition matches a known failure mode, the agent recommends the job plan associated with that failure mode. If the condition does not match a known failure mode, the agent flags it as an emerging issue that may require investigation.

Conversational interface. Through the Maximo AI Assistant, a planner or reliability engineer can ask questions in natural language: "What is happening with this chiller?" The assistant returns a readable answer with the current condition, recent trends, and recommended actions. The conversational interface makes the APM stack accessible to users who are not data scientists or reliability engineers.

Automated work order creation. In the current release, Condition Insight can recommend work orders based on its analysis. The automated or semi-automated creation of those work orders is described as a near-term capability rather than a finished one. The current workflow is that the agent recommends, a reliability engineer reviews, and the work order is created manually or through a workflow rule. The governance of this step is important: an AI recommendation is still a recommendation, and the decision to act on it should be owned by a human.

For teams on MAS 9.x or planning a move to it, there are several steps worth taking before enabling Condition Insight:

1. Rationalize your failure coding. Confirm that failure classes, problem codes, and failure codes are defined, required on closure, and actually being entered. Run a query against the last six months of closed work orders and look at how much content sits in the code structure versus the long description field. If your failure coding is incomplete, the agent cannot map conditions to failure modes effectively.

-- Check failure coding completeness
SELECT
  COUNT(*) AS total_wo,
  SUM(CASE WHEN failurecode IS NOT NULL THEN 1 ELSE 0 END) AS with_failure_code,
  SUM(CASE WHEN problemcode IS NOT NULL THEN 1 ELSE 0 END) AS with_problem_code,
  ROUND(100.0 * SUM(CASE WHEN failurecode IS NOT NULL THEN 1 ELSE 0 END) / COUNT(*), 1) AS pct_coded
FROM WORKORDER
WHERE status = 'CLOSE'
  AND changedate > SYSDATE - 180;

2. Get FMEAs into Maximo. Agentic AI mapping condition to failure modes only works if the failure modes are present as structured data. Even a modest catalog of critical assets with their top failure modes tied to job plans is a meaningful starting point. The Reliability Strategies application in Manage 9.2 includes a built-in FMEA library that makes it easier to create and manage FMEA records.
3. Validate meter coverage on critical assets. Condition Insight reads meter readings. If your critical assets have the right meter groups defined and readings are coming in from the historian or inspection process, you are much closer to a useful pilot. Assets without meter readings will not trigger condition-based recommendations.
4. Agree who owns the output. Decide now whether a Condition Insight suggestion becomes a work request automatically, a triage item for a reliability engineer, or an alert that a planner reviews. That is a governance choice, not a product setting.

Health Scoring: Configuration and Industry Accelerators

Maximo Health calculates asset health scores based on contributors that you define. The scoring model is flexible, but the default configuration covers most use cases. The default scores available out of the box with activation are:

• Health: Based on the number of open service requests, the number of open corrective work orders, and remaining useful life. The health score is a composite metric that reflects the overall condition of the asset.
• Criticality: Normalization of the priority value assigned to the asset. Criticality is used in risk calculation and in prioritizing maintenance work.
• Remaining Useful Life: Calculated as the chronological age of the asset as a proportion of its expected life. This is a straightforward calculation based on the installation date and the expected life defined in the asset record.
• Risk: Calculated as ((100 minus health) divided by 100) multiplied by criticality. Risk is the metric that drives prioritization: a high-criticality asset with low health has high risk and should be prioritized for maintenance.

In MAS 9.2, IBM introduced 66 industry accelerator scoring models for electrical distribution and transmission assets. These accelerators cover five scoring dimensions: Health, Criticality, Effective Age, End of Life Probability, and Risk. The accelerators are pre-built scoring models that you can activate and assign to your assets without configuring the contributors from scratch. This is a significant time saver for utilities and energy companies that manage large fleets of distribution and transmission assets.

Here is an example of how a custom health scoring contributor is configured:

{
  "contributor": {
    "name": "vibration_severity",
    "description": "Vibration reading severity score",
    "dataSource": {
      "type": "meterreading",
      "meterName": "VIBRATION",
      "aggregation": "latest",
      "assetScope": "self"
    },
    "scoring": {
      "method": "threshold",
      "thresholds": [
        {"max": 2.0, "score": 100, "label": "Normal"},
        {"max": 4.0, "score": 70, "label": "Elevated"},
        {"max": 7.0, "score": 40, "label": "High"},
        {"max": 999.0, "score": 10, "label": "Critical"}
      ]
    },
    "weight": 0.3
  }
}

This contributor reads the latest vibration meter reading for the asset and assigns a score based on thresholds. A reading below 2.0 mm/s gets a score of 100 (normal), while a reading above 7.0 mm/s gets a score of 10 (critical). The weight of 0.3 means this contributor contributes 30 percent to the overall health score. You would combine this with other contributors (open work orders, remaining useful life, inspection results) to create the composite health score.

The new Asset and Location Dashboard in MAS 9.2 provides a unified view of asset and location performance with matrix views, automated analysis, work history, and drill-down into sensor and inspection data. The dashboard includes a Reliability Tab with a Failure History card showing a chronological record of asset or location failures and associated work orders, and a Total Maintenance Cost card showing cumulative maintenance expenses. The Alerts and Meters tab provides real-time visibility of condition-based alerts with priority count cards (Urgent, High, Medium, Low) and an alerts table. This dashboard replaces the need to navigate through multiple applications to get a complete picture of an asset's health and maintenance history.

Weibull Analysis and Failure Probability Curves

A new Probability of Failure tab in the Score Settings application in MAS 9.2 brings Weibull analysis directly into Manage. This is a significant addition because Weibull analysis is one of the most powerful tools in reliability engineering, and previously it required exporting data to a separate statistical tool.

Weibull analysis fits a probability distribution to historical failure data and produces a failure curve that shows the probability of failure as a function of time. The Weibull distribution is characterized by two parameters: the shape parameter (beta), which determines the shape of the failure curve, and the scale parameter (eta), which determines the characteristic life. By fitting Weibull parameters from historical work order data directly in Manage, you can visualize failure curves alongside actual failure data and define curves by scoring group.

The shape parameter tells you what kind of failure pattern the asset exhibits:

• Beta less than 1: Infant mortality. The asset is more likely to fail early in its life, and the failure rate decreases over time. This pattern is typical of manufacturing defects or installation errors.
• Beta equal to 1: Random failures. The failure rate is constant, and failures occur randomly over time. This pattern is typical of electronic components or external events.
• Beta greater than 1: Wear-out failures. The failure rate increases over time, and the asset is more likely to fail as it ages. This pattern is typical of mechanical components with wear, fatigue, or degradation.

Here is an example of how Weibull parameters are used to calculate probability of failure:

# Weibull probability of failure calculation
import math

# Parameters from Manage Weibull analysis
beta = 2.3   # shape parameter (wear-out pattern)
eta = 4500   # scale parameter (characteristic life in hours)

# Current age of asset (hours)
current_age = 3200

# Probability of failure by time t
def weibull_cdf(t, beta, eta):
    """Cumulative distribution function for Weibull distribution"""
    return 1 - math.exp(-(t / eta) ** beta)

# Current probability of failure
current_pof = weibull_cdf(current_age, beta, eta)

# Probability of failure in next 500 hours
future_pof = weibull_cdf(current_age + 500, beta, eta) - current_pof

print(f"Asset age: {current_age} hours")
print(f"Current probability of failure: {current_pof:.4f} ({current_pof*100:.1f}%)")
print(f"Probability of failure in next 500 hours: {future_pof:.4f} ({future_pof*100:.1f}%)")

# Output:
# Asset age: 3200 hours
# Current probability of failure: 0.2894 (28.9%)
# Probability of failure in next 500 hours: 0.1103 (11.0%)

With this data directly in Manage, you can set up condition-based maintenance triggers based on the probability of failure rather than on fixed time intervals. A work order can be generated when the probability of failure in the next 30 days exceeds a threshold, which is a more efficient maintenance strategy than a fixed monthly or quarterly inspection.

The Weibull analysis in Manage also supports the Duval triangle for dissolved gas analysis (DGA) in transformer oil. The Duval triangle is a diagnostic tool that maps the relative proportions of three key gases (methane, ethylene, and acetylene) to specific transformer fault types. This is a specialized tool for utilities and power generation companies, and having it directly in Manage eliminates the need for a separate DGA analysis tool.

Reliability Strategies: From FMEA to Living Programs

The Reliability Strategies application in Manage 9.2 has evolved from a static FMEA repository into a dynamic, AI-assisted reliability engineering tool. The built-in FMEA library provides pre-built failure modes for common asset types, which accelerates the initial setup. The AI-assisted strategy development helps reliability engineers build maintenance strategies using natural language prompts, and the validation capabilities ensure that strategies remain relevant as asset conditions change.

The RCM Strategy Builder uses AI to recommend failure modes, mitigation actions, and maintenance tasks based on the asset type, operating context, and historical failure data. This is where the connection to Condition Insight becomes important: Condition Insight maps detected conditions to failure modes, and the failure modes are defined in the Reliability Strategies application. If the failure modes are not defined or are poorly defined, Condition Insight cannot make useful recommendations.

The recommended workflow for setting up reliability strategies is:

1. Start with critical assets. Identify the top 50 or 100 critical assets in your organization. These are the assets where a failure has the highest impact on safety, production, or cost.
2. Use the built-in FMEA library. The library provides common failure modes for pumps, motors, valves, transformers, and other standard asset types. Use these as a starting point and customize them for your specific operating context.
3. Link failure modes to job plans. Each failure mode should have an associated job plan that defines the maintenance work to be performed when that failure mode is detected. This is the link between reliability engineering and maintenance execution.
4. Connect to Condition Insight. Once the failure modes and job plans are in place, Condition Insight can map detected conditions to the failure modes and recommend the corresponding job plans.
5. Review and refine. Reliability strategies are not static. Review them quarterly, update them based on new failure data, and refine the job plans based on the results of the maintenance work.

Practical Implications

For reliability engineers and maintenance strategists, the practical takeaway from MAS 9.2 is that the APM stack has reached a level of integration where you can implement a closed-loop reliability program without stitching together multiple tools. The data flows from sensors to analysis to recommendations to execution and back. The AI layer compresses the analysis time from hours to seconds, and the conversational interface makes the insights accessible to the people who need them.

The investment to get there is not trivial. You need clean data, complete failure coding, FMEA records, meter coverage on critical assets, and a governance model for AI recommendations. But the payoff is significant. IBM reports 47 percent less unplanned downtime and 17 percent longer asset lifespan for organizations using the APM stack with Condition Insight. Whether you achieve those numbers depends on your data quality and your implementation, but the direction is clear: AI-driven reliability is not a future concept. It is available now, and the teams that adopt it will have a measurable advantage over those that do not.

Bottom Line

Maximo APM in 2026 is the most integrated, capable asset performance management platform available, and it is the only one that combines full EAM and APM in a single suite. The MAS 9.2 release, with Condition Insight, industry accelerators, DGA with Duval triangle, and Weibull analysis, provides tools that were previously available only as separate specialized products. The closed-loop architecture from condition monitoring to work order execution, combined with the AI interpretation layer, creates a reliability program that is faster, more consistent, and more scalable than traditional manual analysis. Start with your critical assets, get your failure coding in order, build your FMEA records, and let Condition Insight do the analysis that your reliability engineers have been doing manually. The platform is ready. The question is whether your data is.