Sensor Data Analysis

Reading the Machine's Pulse

Every machine has a heartbeat. A healthy motor hums at a steady frequency. A healthy bearing produces a smooth, low-amplitude vibration signature. A healthy pump maintains consistent pressure and flow. When something changes — even slightly — the data tells the story long before a human can hear, feel, or see the problem.

The challenge is not collecting data. Modern sensors generate thousands of readings per second. The challenge is making sense of it all. A single vibration sensor sampling at 10 kHz produces 864 million data points per day. No engineer can review that manually. This is where AI becomes essential — not as a replacement for engineering knowledge, but as a tireless analyst that never blinks, never takes a break, and never misses a trend.

This chapter teaches you to think about sensor data the way AI does — recognizing patterns, understanding baselines, and identifying the signatures of common failure modes. You will not write code, but you will learn to interpret what the AI tells you and validate its conclusions with your engineering experience.

Types of Sensor Data in Engineering

Vibration Data

Vibration is the single most informative measurement for rotating equipment. It reveals problems in bearings, gears, shafts, couplings, and structural components — often weeks or months before failure.

What vibration sensors measure:

Key insight: Different faults produce vibration at different frequencies. A shaft imbalance vibrates at 1x running speed. Misalignment shows at 2x. A bearing defect produces vibration at very specific frequencies determined by bearing geometry. AI models learn these frequency signatures and match them to known fault patterns.

Normal vibration levels (ISO 10816 guidelines):

Open data/vibration-data.csv in the code panel. This file contains 60 days of daily vibration readings from a 45 kW motor driving a centrifugal pump. Look at the overall trend — the first 40 days show stable readings around 1.5 mm/s (acceptable zone). From Day 41, you will see a gradual upward trend reaching 3.8 mm/s by Day 58. The AI flagged this on Day 45. The bearing was replaced on Day 55. Without AI, this motor would have failed catastrophically around Day 65-70.

Temperature Data

Temperature is the second most common monitoring parameter. It is simpler to measure than vibration but often provides later warning — by the time temperature rises significantly, damage may already be occurring.

What temperature reveals:

Normal temperature ranges (general guidelines):

Indian context: Ambient temperature in Indian factories can reach 45-50 degrees C in summer (especially in non-air-conditioned plants in Rajasthan, Gujarat, Vidarbha). This means temperature thresholds must be seasonally adjusted. An AI model trained on annual data automatically accounts for this — a fixed threshold would generate false alarms every summer.

Open data/temperature-logs.csv to see 90 days of hourly temperature readings from a transformer at an industrial substation. Notice the daily cycle (lower at night, higher during peak load hours) and the seasonal trend. Around Day 72, you will see the night-time baseline start rising — the cooling fans were partially blocked by dust accumulation. The AI detected this deviation from the learned pattern three days before the high-temperature alarm would have triggered.

Pressure Data

Critical for hydraulic systems, pneumatic circuits, pumps, and process piping.

What pressure patterns reveal:

Current Data (Motor Current Signature Analysis)

Every motor draws current proportional to its mechanical load. Changes in current patterns reveal both electrical and mechanical problems without needing to install vibration sensors.

What current analysis reveals:

What Normal Looks Like vs Anomaly

Establishing a Baseline

Before AI can detect anomalies, it must learn what "normal" looks like. This is not a single number — it is a pattern that varies by:

AI models typically need 2-4 weeks of "healthy" data to establish a reliable baseline. During this learning period, the engineer must confirm the machine is actually healthy — otherwise the AI learns an already-degraded state as "normal."

Types of Anomalies

Trend anomaly — a gradual, consistent change in one direction. Example: vibration increasing by 0.05 mm/s per day. This indicates progressive degradation (bearing wear, imbalance worsening).

Step change — a sudden jump to a new level that stays there. Example: temperature jumps from 65 to 78 degrees C overnight. This indicates a discrete event (loss of one cooling fan, sudden load increase, lubrication change).

Spike anomaly — brief, sharp deviations that return to normal. Example: pressure spikes lasting 2-3 seconds every few hours. This indicates intermittent issues (valve sticking, cavitation, foreign objects in flow).

Pattern change — the overall shape of the data changes even if absolute values look acceptable. Example: vibration spectrum shifts its dominant frequency from 1x to 2x running speed. This indicates a change in fault mechanism (from imbalance to misalignment).

Vibration Analysis Basics

Imbalance

The most common fault in rotating machinery. An uneven mass distribution causes the shaft to vibrate at exactly 1x running speed (e.g., a motor at 1480 RPM shows dominant vibration at 24.7 Hz). The amplitude is proportional to the severity. Fix: balance the rotor.

Misalignment

The second most common fault. Angular or parallel offset between coupled machines causes vibration at 2x running speed (sometimes 3x). Also produces high axial vibration. Fix: realign using laser alignment tools.

Bearing Wear

As bearings degrade, they produce vibration at very specific frequencies determined by bearing geometry (inner race, outer race, ball/roller, cage frequencies). Early wear shows as subtle high-frequency noise. Advanced wear shows as clear peaks at bearing defect frequencies. Fix: replace bearing before catastrophic failure.

Looseness

Mechanical looseness (loose bolts, cracked frame, worn fits) produces vibration at many harmonics of running speed (1x, 2x, 3x, 4x, etc.) and often sub-harmonics (0.5x). The spectrum looks "messy" with many peaks. Fix: tighten, replace worn components, or re-machine fits.

How AI Spots Patterns Humans Miss

A skilled vibration analyst can review perhaps 20-30 machines per day, spending 10-15 minutes per machine examining spectra and trends. An AI system monitors all machines continuously and can detect:

Key Takeaways