Understanding Prognosis in Predictive Maintenance

Devices

Portable balancer & Vibration analyzer Balanset-1A

$2,353.83

Parts

Vibration sensor

$107.26

Parts

Optical Sensor (Laser Tachometer)

$147.78

Devices

Balanset-4

$8,107.90

Parts

Magnetic Stand Insize-60-kgf

$54.82

Parts

Reflective tape

$11.92

Devices

Dynamic balancer “Balanset-1A” OEM

$2,067.80

Definition: What is Prognosis?

Prognosis (also called remaining useful life estimation or failure prediction) is the process of estimating how much time remains before a detected fault will cause functional failure or require intervention. Prognosis follows fault detection (identifying a problem exists) and diagnosis (identifying what the problem is), answering the critical question “When must we act?” through analysis of vibration progression trends, fault type characteristics, and equipment operating conditions.

Accurate prognosis is what makes predictive maintenance truly predictive—it enables scheduling maintenance at the optimal time (neither too early, wasting remaining life, nor too late, risking failure), procurement of long-lead parts, resource allocation, and production scheduling coordination.

Prognostic Methods

1. Trend Extrapolation

Most common and practical method:

• Plot historical vibration data vs. time
• Fit trend line (linear, exponential, etc.)
• Extrapolate to predict when alarm or failure threshold crossed
• Update prediction with each new measurement
• Accuracy: Moderate (assumes trend continues)
• Requirements: Sufficient trending history (6+ data points minimum)

2. Physics-Based Models

• Use understanding of failure physics (crack growth, spall propagation)
• Model predicts progression based on stress, cycles, environment
• Example: Paris Law for crack growth, bearing L10 life calculations
• Accuracy: Good if model parameters known
• Requirements: Detailed equipment and operating data

3. Experience-Based (Historical Data)

• Based on past failures of similar equipment
• Typical progression rates from history
• Empirical relationships (vibration level → time to failure)
• Accuracy: Fair, equipment-specific
• Requirements: Historical failure database

4. Statistical/Machine Learning

• Train algorithms on historical progression data
• Pattern recognition from many similar cases
• Probabilistic predictions
• Accuracy: Can be very good with sufficient data
• Requirements: Large dataset, computational resources

Factors Affecting Prognosis Accuracy

Trending Data Quality

• More data points → better trend definition
• Consistent measurements → reliable trends
• Adequate history (months minimum)
• Clean data (outliers identified)

Fault Progression Characteristics

• Predictable Progression: Easier to forecast (gradual bearing wear)
• Accelerating Progression: Harder (bearing spall growth exponential)
• Erratic Progression: Difficult (looseness, intermittent rubs)
• Sudden Failures: Unpredictable (shaft fracture from crack)

Operating Condition Stability

• Stable conditions → reliable predictions
• Variable loads/speeds → predictions less certain
• Process changes can accelerate or slow progression

Remaining Useful Life (RUL) Estimation

Definition

• Time from current condition to failure or intervention threshold
• Expressed in operating hours, days, or calendar time
• Updated continuously as new data collected

Confidence Intervals

• RUL is estimate with uncertainty
• Express as range (30-90 days with 90% confidence)
• Uncertainty decreases as failure approaches (more data)
• Conservative estimates for critical equipment

Example

• Bearing defect detected at 2g envelope amplitude
• Historical progression: 2g → 10g (alarm level) in 60 days typical
• Current rate: increasing 0.5g per week
• Prediction: Alarm level in ~10 weeks
• Recommendation: Schedule maintenance within 6-8 weeks

Applications

Maintenance Scheduling

• Plan outage when RUL indicates optimal timing
• Coordinate with production schedules
• Group repairs to minimize downtime
• Avoid both premature and late interventions

Parts Management

• Order spare parts with appropriate lead time
• Avoid expediting costs
• Reduce safety stock requirements
• Just-in-time provisioning based on prognosis

Resource Allocation

• Prioritize among multiple degrading machines
• Allocate limited resources to most urgent needs
• Plan workforce assignments
• Tool and equipment staging

Challenges and Limitations

Prediction Uncertainty

• Fault progression not perfectly predictable
• Operating conditions may change
• Unexpected accelerations possible
• Always maintain safety margins

Data Requirements

• Need adequate trending history
• Early in fault development, predictions less certain
• Improved as more data collected

Multiple Failure Modes

• Predicting one mode while another causes failure
• Comprehensive monitoring helps
• Must consider all active degradation mechanisms

Improving Prognostic Accuracy

Increase Measurement Frequency

• More data points → better trend definition
• Detect acceleration earlier
• Reduce prediction uncertainty

Multiple Parameters

• Vibration + temperature + oil analysis
• Corroborating indicators improve confidence
• Different parameters may have different lead times

Continuous Updating

• Revise prognosis with each new measurement
• Don’t rely on single early prediction
• Adapt to actual progression rate

Prognosis is the predictive element that distinguishes true predictive maintenance from simple condition monitoring. By estimating remaining useful life and failure timelines from trending data and fault progression understanding, prognosis enables the optimized maintenance timing that maximizes equipment utilization while maintaining reliability—the ultimate goal of condition-based maintenance strategies.

---

← Back to Main Index