AI Equipment Vibration and Temperature Monitoring System

Development of an AI system for monitoring equipment vibration and temperature

Vibration and temperature are two key diagnostic parameters for rotating equipment. Vibration provides information about mechanical defects (bearings, imbalance, misalignment), while temperature provides information about thermal anomalies (bearing overheating, lubrication problems, overload).

Placement of sensors

Vibration sensors:

sensor_placement_guidelines = {
    'motor_drive_end_bearing': {
        'position': 'горизонтально на крышке подшипника',
        'sensitive_to': ['unbalance', 'misalignment', 'bearing_defects'],
        'frequency_range': '10-10000 Hz'
    },
    'motor_non_drive_end_bearing': {
        'position': 'горизонтально',
        'sensitive_to': ['rotor_asymmetry', 'bearing'],
        'frequency_range': '10-10000 Hz'
    },
    'pump_bearing': {
        'position': 'на корпусе насоса у подшипника',
        'sensitive_to': ['cavitation', 'impeller_unbalance'],
        'note': 'добавить радиальный + осевой датчики'
    },
    'gearbox': {
        'position': 'на корпусе редуктора',
        'sensitive_to': ['gear_mesh_frequency', 'tooth_defects'],
        'frequency_range': '10-20000 Hz'
    }
}

Thermal sensors:

• PT100 / PT1000: accuracy ±0.1°C, for bearings < 2 seconds response
• Thermocouple: Faster response, suitable for hot spots
• Infrared thermometers: non-contact, for periodic inspections
• Thermal imaging: periodic inspection, sees "hot spots"

Vibration signal processing

ISO 10816 diagnostic zones:

import numpy as np

def classify_vibration_severity(rms_velocity_mm_s, machine_class):
    """
    ISO 10816: классификация виброскорости по зонам A/B/C/D
    Класс I: малые машины < 15 кВт
    Класс II: средние машины 15-75 кВт
    Класс III: крупные машины > 75 кВт на жёстком основании
    """
    thresholds = {
        'I':   {'A': 0.28, 'B': 0.71, 'C': 1.8, 'D': float('inf')},
        'II':  {'A': 0.45, 'B': 1.12, 'C': 2.8, 'D': float('inf')},
        'III': {'A': 0.71, 'B': 1.8,  'C': 4.5, 'D': float('inf')},
        'IV':  {'A': 1.12, 'B': 2.8,  'C': 7.1, 'D': float('inf')}
    }

    t = thresholds[machine_class]
    if rms_velocity_mm_s <= t['A']:
        return 'A', 'Excellent — new machine condition'
    elif rms_velocity_mm_s <= t['B']:
        return 'B', 'Acceptable — long-term operation allowed'
    elif rms_velocity_mm_s <= t['C']:
        return 'C', 'Tolerable — short-term only, schedule maintenance'
    else:
        return 'D', 'Unacceptable — immediate shutdown risk'

Spectral analysis – automatic detection of defective frequencies:

from scipy.fft import fft, fftfreq
import scipy.signal as signal

def diagnose_from_spectrum(vibration_signal, sampling_rate, shaft_rpm,
                            bearing_freqs, gear_mesh_freq=None):
    """
    Автоматическая диагностика по спектру вибрации
    """
    n = len(vibration_signal)
    freqs = fftfreq(n, 1/sampling_rate)[:n//2]
    magnitude = np.abs(fft(vibration_signal))[:n//2]

    shaft_freq = shaft_rpm / 60
    diagnoses = []

    # Дисбаланс: доминирование 1x оборотной частоты
    idx_1x = np.argmin(np.abs(freqs - shaft_freq))
    if magnitude[idx_1x] > np.mean(magnitude) * 10:
        diagnoses.append({'fault': 'unbalance', 'severity': 'medium',
                          'evidence': f'1x amplitude = {magnitude[idx_1x]:.2f}'})

    # Расцентровка: высокие 2x + осевая составляющая
    idx_2x = np.argmin(np.abs(freqs - 2 * shaft_freq))
    if magnitude[idx_2x] > magnitude[idx_1x] * 0.5:
        diagnoses.append({'fault': 'misalignment', 'severity': 'medium',
                          'evidence': f'2x/1x ratio = {magnitude[idx_2x]/magnitude[idx_1x]:.2f}'})

    # Подшипниковые дефекты: BPFO, BPFI
    for bearing_name, bf in bearing_freqs.items():
        for fault_name, freq in [('bpfo', bf['bpfo']), ('bpfi', bf['bpfi'])]:
            idx = np.argmin(np.abs(freqs - freq))
            local_mean = magnitude[max(0, idx-10):idx+10].mean()
            if magnitude[idx] > local_mean * 5:
                diagnoses.append({
                    'fault': f'{fault_name}_{bearing_name}',
                    'severity': 'high',
                    'evidence': f'Amplitude at {freq:.1f}Hz = {magnitude[idx]:.2f}'
                })

    return diagnoses

Temperature monitoring

Baseline and anomalies:

class TemperatureMonitor:
    def __init__(self, baseline_window_hours=24):
        self.baseline = {}
        self.alert_thresholds = {}

    def set_baseline(self, equipment_id, temperature_history):
        """
        Baseline: медианная температура в нормальном режиме
        Учитывает нагрузку: температура нормальна при высокой нагрузке
        """
        self.baseline[equipment_id] = {
            'overall_median': np.median(temperature_history),
            'load_corrected': self._fit_load_temperature_curve(temperature_history)
        }

    def check_temperature(self, equipment_id, current_temp, current_load):
        baseline = self.baseline[equipment_id]
        expected_temp = baseline['load_corrected'].predict([[current_load]])[0]

        deviation = current_temp - expected_temp

        if deviation > 20:
            return 'CRITICAL', f'Temperature {deviation:.1f}°C above expected'
        elif deviation > 10:
            return 'WARNING', f'Temperature {deviation:.1f}°C above expected'
        elif deviation > 5:
            return 'CAUTION', f'Slight temperature elevation: {deviation:.1f}°C'
        else:
            return 'NORMAL', None

Temperature trend analysis:

def analyze_temperature_trend(temperature_series, window_days=7):
    """
    Постепенный рост температуры = деградация подшипника или нарушение смазки
    """
    recent = temperature_series.last(f'{window_days}D')
    trend = np.polyfit(range(len(recent)), recent.values, 1)[0]

    rate_per_day = trend
    if rate_per_day > 1.0:  # > 1°C в день = красный флаг
        return 'ACCELERATING', rate_per_day
    elif rate_per_day > 0.3:
        return 'GRADUAL_RISE', rate_per_day
    else:
        return 'STABLE', rate_per_day

Correlation of vibration and temperature

Multi-sensor fusion:

def correlate_vibration_temperature(vibration_features, temperature_features, time_lag_hours=2):
    """
    Тепловые признаки могут запаздывать за вибрационными на 1-4 часа
    (тепло распространяется медленнее вибрации)
    """
    # Lag correlation
    combined_score = 0

    # Оба показателя выше нормы = усиленный сигнал
    if vibration_features['kurtosis'] > 3 and temperature_features['deviation'] > 5:
        combined_score = max(
            vibration_features['anomaly_score'],
            temperature_features['anomaly_score']
        ) * 1.3  # усиление при подтверждении двумя сенсорами

    # Температура без вибрации = возможно смазки/охлаждение
    elif temperature_features['deviation'] > 10 and vibration_features['kurtosis'] < 2:
        return 'lubrication_or_cooling_issue'

    return combined_score

Dashboard and alerting

Real-time Grafana dashboard:

• Live waveform: last 10 seconds of vibration signal
• FFT spectrum: updated every 5 minutes
• Trending: RMS and kurtosis in 30 days
• ISO zone: color indication A/B/C/D
• Temperature heatmap: all measurement points on the equipment diagram

Alert Matrix:

alert_matrix = {
    ('D', 'CRITICAL'): 'immediate_shutdown',  # ISO зона D + критическая температура
    ('D', 'WARNING'): 'urgent_inspection',
    ('C', 'CRITICAL'): 'urgent_inspection',
    ('C', 'WARNING'): 'schedule_next_shift',
    ('B', 'CAUTION'): 'monitor_closely',
    ('A', 'NORMAL'): 'routine'
}

Deadlines: Sensors + OPC-UA/Modbus + basic vibration RMS + ISO classification + temperature alerts + Grafana — 3-4 weeks. Spectral diagnostics, envelope analysis, temperature trend analysis, multi-sensor fusion, mobile alerts — 2-3 months.