32 Predictive Maintenance: Models and Rollout

32.5.1 Thermal Monitoring Architecture

Figure 32.2 answers what the camera alone cannot: how a hot reading turns into an action.

Read Figure 32.2 downward in three moves. The top band is sensing, and it already forces a choice: a handheld camera suits an inspection route, a fixed sensor suits repeatable trending, and a drone survey suits assets nobody can reach safely. The middle band is where a temperature becomes evidence, because a reading only means something next to a baseline and the rise above ambient. That comparison sorts a site into normal, warning and critical. The bottom band then aims the effort at bus bars, bearings or heat exchangers. The same three steps serve all three families, so the workflow is reusable even when the failure physics differ.

Thermal monitoring architecture showing infrared sensing technologies feeding into analysis engine for baseline comparison, trending, and anomaly detection across electrical, mechanical, and process equipment applications.

32.5.2 Electrical Applications

• Hot spots on connections indicate high resistance
• Overheated components indicate overload
• Phase imbalance in motors
• Can detect problems 6-12 months in advance

32.5.3 Mechanical Applications

• Bearing overheating (friction)
• Belt misalignment (heat buildup)
• Lubrication issues (dry bearings)
• Coupling problems

32.5.4 Temperature Thresholds

32.6.1 ML Model Selection Decision Tree

Figure 32.3 sorts model families by the labels you hold, not by the algorithm you like.

Begin at the question box in Figure 32.3, which asks what the maintenance team needs to know. The three branches below answer it in the order of evidence you can supply. Supervised learning comes first, and it needs labelled failures, which many sites do not have. Unsupervised learning needs only a record of normal running, so it suits a plant with clean history and few recorded faults. Forecasting sits last because it needs runs that were followed all the way to failure, the hardest data of all. Each branch also states its own question: will this asset fail soon, is this pattern odd now, how long is left.

Use this decision tree to select the appropriate ML approach based on your available data and prediction goals.

32.6.2 Supervised Learning

Approach: Requires labeled failure data to train classifiers.

Algorithms:

• Random Forest, XGBoost for classification
• Neural networks for complex patterns

Output: “Will this bearing fail in next 30 days?” (Yes/No with probability)

Requirements:

• Historical failure data (dozens to hundreds of examples)
• Consistent sensor data leading up to failures
• Domain expertise to label failure modes

32.6.3 Unsupervised Learning

Approach: Learns normal operation without failure labels.

Algorithms:

• Autoencoders (reconstruction error indicates anomaly)
• Isolation Forests (detects outliers)
• One-class SVM

Output: “Is this vibration signature abnormal?” (Anomaly score)

Advantages:

• Works without historical failures
• Detects novel failure modes
• Good for rare events

32.6.4 Time-Series Forecasting

Use this time-series forecasting section as a guided decision record, not as a list to memorise. First identify the stated input, assumption, or scenario; then compare each option on the same units and time boundary. Next check which value changes the outcome and which evidence would reveal an invalid assumption. For time-series forecasting, the useful result is the reasoning chain: observed condition, governing constraint, calculation or classification, and operational consequence. Record that chain before choosing an answer or carrying a value into the next section. Where the panel supplies several choices, reject each distractor against the chapter’s named mechanism instead of relying on wording cues. Where it supplies a table or timeline, compare rows at like-for-like scale and preserve the difference between an early indication, an actionable threshold, and a final outcome. This turns time-series forecasting into evidence that can be reviewed, recalculated, and connected to the running design narrative.

Approach: Predicts remaining useful life (RUL) based on degradation trends.

Algorithms:

• LSTM neural networks
• Prophet (trend + seasonality)
• Gaussian Process Regression

Output: “How many hours/days until failure?” (RUL estimate with confidence interval)

Key metrics:

• Mean Absolute Error (MAE)
• Root Mean Square Error (RMSE)
• Percentage within 10%/20% tolerance

32.8.1 Cost Components

Investment costs (illustrative ranges; replace with local quotes):

• Sensors: $100-500 per motor (vibration, temperature)
• Gateways: $500-2,000 per zone
• Software: $50,000-500,000 (depending on scale)
• Integration: 2-5x hardware cost for brownfield
• Training: $1,000-5,000 per technician

Operating costs (illustrative ranges; replace with local contracts):

• Platform licensing: $10-50 per asset/month
• Connectivity: $5-20 per gateway/month
• Data storage: $0.02-0.05 per GB/month
• Analyst time: $50,000-100,000/year for dedicated resources

32.8.2 Benefit Categories

Direct savings:

• Reduced emergency repairs (labor + parts + expediting)
• Extended equipment life (deferred replacement)
• Lower spare parts inventory (order when needed)
• Reduced energy consumption (efficient equipment)

Indirect savings:

• Avoided production losses (unplanned downtime)
• Improved quality (equipment in specification)
• Reduced safety incidents (early warning of hazards)
• Better capital planning (known equipment condition)

32.8.3 Sample ROI Calculation

Use this sample roi calculation section as a guided decision record, not as a list to memorise. First identify the stated input, assumption, or scenario; then compare each option on the same units and time boundary. Next check which value changes the outcome and which evidence would reveal an invalid assumption. For sample roi calculation, the useful result is the reasoning chain: observed condition, governing constraint, calculation or classification, and operational consequence. Record that chain before choosing an answer or carrying a value into the next section. Where the panel supplies several choices, reject each distractor against the chapter’s named mechanism instead of relying on wording cues. Where it supplies a table or timeline, compare rows at like-for-like scale and preserve the difference between an early indication, an actionable threshold, and a final outcome. This turns sample roi calculation into evidence that can be reviewed, recalculated, and connected to the running design narrative.

Scenario: 100-motor manufacturing plant using editable planning assumptions.

With predictive maintenance:

The table’s $180,000 investment is a separate case from the interactive calculator’s $140,000 default; the calculator does not reproduce this table unless its investment inputs are aligned.

html`<div style="margin: 20px 0;">
<h4 style="color: #2C3E50;">Investment Breakdown</h4>
${Plot.plot({
marginLeft: 100,
x: {label: "Cost ($)", grid: true},
y: {label: null},
marks: [
Plot.barX(roiBreakdown, {
y: "category",
x: "cost",
fill: "#3498DB",
tip: true,
title: d => `${d.category}: $${d.cost.toLocaleString()} (${d.percent}%)`
}),
Plot.text(roiBreakdown, {
y: "category",
x: "cost",
text: d => `$${(d.cost/1000).toFixed(0)}K`,
dx: -30,
fill: "white",
fontSize: 12,
fontWeight: "bold"
}),
Plot.ruleX([0])
]
})}
</div>`

32.9.1 Phase 1: Pilot (Months 1-6)

• Select 10-20 critical assets
• Deploy basic vibration and temperature sensors
• Establish data collection infrastructure
• Create baseline normal operation profiles
• Success metric: Detect one previously undetected issue

32.9.2 Phase 2: Expansion (Months 7-18)

• Expand to 50-100 assets
• Implement ML-based anomaly detection
• Integrate with CMMS for work order generation
• Train maintenance technicians on new tools
• Success metric: measured reduction in emergency work on pilot asset classes

32.9.3 Phase 3: Optimization (Months 19-36)

• Full facility coverage (all critical assets)
• Remaining useful life predictions
• Automated parts ordering
• Continuous model improvement
• Success metric: sustained planned-maintenance ratio and technician-confirmed alert quality