Predictive Facilities Maintenance

Define the building service that maintenance must protect

Facilities maintenance supports outcomes such as thermal comfort, indoor air quality, water availability, vertical transport, security, fire protection and continuity of critical spaces. An asset matters because of the service it provides, not because it appears in the asset register.

Marketways links each plant item, component, sensor and control point with the zones, systems and occupants that depend on it. We record available redundancy and the consequence of losing it. This service map establishes why a developing fault matters and which person has authority to review the response.

Use a different maintenance question for each system

Building systems do not deteriorate in one uniform way. A bearing may develop a vibration signature, a filter may accumulate pressure loss, a valve may leak while appearing closed and a control sequence may be wrong from the day it is commissioned. A life-safety system can also require statutory inspection even when condition data show no anomaly.

Marketways identifies the failure mode and the decision before choosing a method. Condition monitoring may support an inspection threshold. Automated fault detection and diagnostics may identify an operating deviation and its likely cause. Reliability analysis may estimate failure risk, while statutory requirements determine mandatory work. One generic health score cannot replace these different questions.

Establish how normal operation changes with context

Outdoor temperature, humidity, occupancy, operating hours, setpoints and building use change the signals produced by heating, cooling and ventilation systems. High energy use on a hot, occupied afternoon may be normal. The same use overnight may indicate poor scheduling, simultaneous heating and cooling or an override that nobody removed.

Marketways builds a baseline for defined operating states instead of comparing every observation with one average. The model distinguishes seasonal and operational variation from behaviour that remains unusual under comparable conditions. Facilities engineers can then examine an anomaly that has a clear context rather than another unexplained alert.

Verify sensors, point names and control logic before modelling

A predictive model inherits the quality of the building data it receives. Sensors can drift or freeze. Point names can differ between buildings, and an automation-system upgrade can change units, sampling or control sequences without changing the label.

Marketways checks ranges, units, timestamps, missing periods, calibration evidence and relationships between connected points. We align equipment and sensor identities through a consistent building-system model. A model should not diagnose a failing chiller when the real defect is a mislabelled temperature point.

Distinguish a fault from its symptom

One physical or control fault can create several symptoms across an interconnected building system. A stuck cooling-coil valve may alter supply temperature, fan demand, zone comfort and energy consumption. Treating each signal as an independent fault can create duplicate work orders and conceal the common cause.

Marketways represents plausible cause-and-effect relationships between components, controls and observed states. We combine engineering rules, statistical evidence and system topology to narrow the diagnosis. The output states the detected condition, credible causes and evidence still required; it does not present an uncertain diagnosis as an established physical fact.

Maintenance history contains earlier policy choices

Work orders do not provide a neutral record of every fault. Teams inspect critical or troublesome assets more often, preventive schedules create observations at fixed intervals and some failures disappear when a contractor replaces a component without recording the cause. Assets with sparse records may be reliable or simply neglected.

Marketways examines inspection coverage, maintenance policy, unrecorded downtime, repeated temporary repairs and changes in coding practice. We distinguish the event the model should predict from the decision that made the event visible. This step prevents a model from learning yesterday's maintenance priorities as though they were the natural failure process.

Define an intervention window, not merely a risk score

A warning has operational value only when the facilities team has time to inspect, plan access, obtain a part and complete the work before the consequence becomes unacceptable. A prediction that arrives after the safe maintenance window has closed cannot prevent the disruption.

Marketways estimates the lead time available for each credible failure mode and compares it with the organisation's response time. The resulting trigger reflects the decision horizon. A slowly degrading pump and an unstable life-safety control therefore require different warning rules even when their modelled failure probabilities are similar.

Prioritise faults through service consequence and feasibility

An energy-cost estimate can help rank building faults, but cost alone does not represent occupant harm, critical service, safety or system interaction. A minor fault in a redundant comfort system may wait. A less expensive fault affecting ventilation in a critical space may require immediate qualified review.

Marketways combines probability, service consequence, affected zones, redundancy, energy and water impact, escalation risk and feasible response. Mandatory technical or safety findings remain outside discretionary trade-offs. The priority tells the facilities team why to act, what must be protected and which constraint limits the response.

Match the alert with labour, access and materials

A technically correct diagnosis does not create a repair slot. The building may require a permit, tenant access, a shutdown window, a specialist contractor, temporary equipment or a spare part with a long lead time.

Marketways connects predicted work with skills, rosters, contracts, access and inventory. We test whether the proposed intervention can be completed within the available window and prepare an alternative when it cannot. The maintenance plan becomes executable instead of sending urgent work into a queue that has no capacity to deliver it.

Control the burden created by false alarms

A model that detects nearly every fault may still fail in practice if it overwhelms facilities staff with weak alerts. Repeated false alarms consume inspection time and teach operators to ignore the system. A missed critical fault, however, can carry a much larger consequence than several unnecessary inspections.

Marketways evaluates precision, recall, lead time and calibration by failure mode and consequence. We also measure alerts per technician, duplicate alerts and the proportion that lead to a useful action. Thresholds reflect the operating cost of both false alarms and missed failures rather than maximising one generic accuracy score.

Validate across seasons, buildings and operating regimes

A model trained during one season or in one building may confuse a new operating regime with failure. Refurbishment, changes in occupancy, revised setpoints and equipment replacement can also invalidate the historical baseline.

Marketways tests performance across relevant weather, load, site and equipment conditions. We hold out complete time periods or buildings where that design reflects intended use. Facilities teams receive limits stating where the model has evidence, where it requires closer review and which changes should trigger recalibration.

An air-handling unit shows the complete diagnostic chain

Suppose an air-handling unit begins using more cooling energy while several zones report discomfort. The increased consumption does not identify whether the cause is weather, occupancy, a leaking valve, sensor drift, poor scheduling or an incorrect control sequence.

Marketways first compares the unit with its normal behaviour under similar conditions. We then test related temperatures, valve commands, fan states, zone demand and work history against the system topology. If the evidence supports a valve problem, the facilities team can schedule a physical inspection during an available access window. The team records the finding and verifies whether repair restores control, comfort and energy performance.

Agentic AI can coordinate an approved maintenance response

Once a diagnostic rule or model identifies a material condition, staff may need to retrieve drawings, check warranty and contract terms, locate a spare, find an operating window and prepare a work order. These steps create an execution gap between the analytical signal and maintenance action.

An Agentic AI workflow can assemble the approved evidence, create a permitted task, request missing information and monitor completion. Marketways tests tool permissions, asset identity, stale documents, conflicting diagnoses and interrupted cases. The workflow cannot approve a safety disposition, alter a control sequence or close a technically disputed fault without the required authority.

Verify the repair and learn from the disposition

Closing a work order confirms an administrative action, not that the predicted fault existed or that the repair restored the service. Without technical disposition, false diagnoses remain labelled as successes and recurring faults return to the queue.

Marketways records what engineers found, which component or setting changed and how the relevant signals behaved afterwards. We measure recurrence, emergency work, availability, comfort, energy, response time and maintenance cost where each outcome is relevant. The disposition improves future diagnosis and reveals when the original model no longer represents the building.

Scale through shared logic without assuming identical buildings

A property portfolio benefits from common asset definitions, failure modes, evidence requirements and review rules. However, identical equipment can serve different spaces, use different control sequences and face different access or redundancy conditions.

Marketways separates reusable diagnostic logic from site-specific configuration. We test the mapping and baseline at each property before enabling portfolio comparison. This structure reduces repeated analytical effort while preserving the local context that determines whether a fault is material.

How Marketways designs predictive facilities maintenance

Marketways combines Predictive Maintenance and Reliability, Risk Detection, engineering evidence, reliability analysis, statistical modelling and AI workflow design. We define failure modes, validate building data, estimate warning windows, prioritise service consequences and connect each accepted signal with an executable maintenance response.

The result is a controlled maintenance system that states which condition requires attention, why it matters, how certain the diagnosis is, when the team must act, which constraints affect delivery and whether the completed intervention restored the required building service.

Related guides and evidence

• Building Energy and Operational Performance
• Tenant, Resident and Community Experience
• Portfolio Capital and Refurbishment Prioritisation
• Airport, Hotel and Attraction Asset Reliability
• Predictive Maintenance and Spares
• Field Workforce and Spares Readiness
• Predictive Maintenance with AI Agents
• Property Portfolio Maintenance, Energy and Capital Prioritisation: case study