AI Status — Intelligent Inspections

AI Engine Status

Total Models

Active (Heuristic)

Trained (Neural)

Training Data

215

LSTM / RandomForest Classifier

Time-series failure prediction — RandomForest replaces LSTM heuristic when trained

Status Trained (RandomForest)

Mode sklearn RandomForest — 253 features

Last Run 2026-04-19

Confidence Trained

Details Top features: total_length_m: 0.031, lidar_bend_length_m: 0.024, hot_spot_ratio: 0.023, mean_temp_f: 0.021, thermal_gradient: 0.021

Transformer Anomaly Detector

Pattern recognition across multi-sensor streams

Status Trained (Neural)

Mode Neural attention weights active

Last Run 2026-04-19

Confidence N/A

Graph Neural Network (GNN)

Cascading failure path modeling between components

Status Trained (Neural)

Mode Learned edge weights + decay active

Last Run 2026-04-19

Confidence Trained

Autoencoder / IsolationForest

Unsupervised anomaly detection — IsolationForest replaces Autoencoder heuristic when trained

Status Trained (IsolationForest)

Mode sklearn IsolationForest — anomaly scoring

Last Run 2026-04-19

Confidence Trained

Reinforcement Learning Scheduler

Maintenance schedule optimization and cost-benefit analysis

Status Q-Table Active (117 states)

Mode Q-learning — 117 state entries

Last Run 2026-04-19

Confidence Active

Federated Learning

Distributed model updates across inspection sites

Status Awaiting Sites

Mode No site updates received

Last Run N/A

Confidence N/A

Physics Validator

Cross-checks sensor outputs against known material behavior and mechanical limits

Status Active — Rule-based

Mode Einstein/Newton/Tesla proxy validation

Last Run 2026-04-19

Confidence Always Active

Details Threshold: 70% confidence trigger

4 model(s) trained from 215 historical inspections using scikit-learn (RandomForest, IsolationForest). Remaining models use rule-based heuristics calibrated to industrial standards. Last inspection: 2026-04-19.