AI Predictive Maintenance System for IoT Devices

A fleet of 10,000 smart meters. One failure means a month of lost data. A technician visit costs are determined per project, but 70% of calls are false alarms. The classic approach—scheduled maintenance or reactive response—is inefficient: either you miss failures or waste resources. We built an ML system that predicts failures 30 days in advance with 88% recall and reduces unplanned downtime by 30%. Over 5 years, we have implemented over 50 projects in predictive maintenance for IoT—from industrial sensors to medical equipment. The system processes telemetry in real time, detects anomalies, and generates a risk score for each device. The dispatch algorithm optimally allocates resources, reducing technician visits by 40%.

How AI Solves the Class Imbalance Problem

Failures account for 0.1–2% of observations. Standard logistic regression gives recall < 30%. We use LightGBM with class_weight='balanced' and population training: first train a global model on the entire fleet, then fine-tune per device. This yields 88% recall even on rare failures. For LSTM RUL, which requires more data, we apply synthetic failure augmentation (SMOTE).

How We Build the Predictive Maintenance Pipeline

Device → MQTT Broker (EMQX) → Kafka → InfluxDB/TimescaleDB
                                     ↓
                              ML Pipeline (Kubeflow)
                                     ↓
                              Model Registry (MLflow)
                                     ↓
                              Triton Inference Server

Stack: MQTT for telemetry collection, Kafka for buffering, InfluxDB for time-series storage. ML pipeline on Kubeflow, experiments logged in MLflow. Inference via Triton Inference Server with ONNX Runtime. For edge devices—quantized ONNX Runtime (4x model size reduction with <1% accuracy loss).

LightGBM is easier to deploy (4 weeks). LSTM for critical systems where RUL accuracy matters.

Why LightGBM Wins on Speed and Interpretability

For tasks with 0.1% failures, LightGBM delivers stable 88% recall without complex tuning. Inference on CPU takes 0.3 ms—enabling predictions on every telemetry message in real time. Unlike LSTM, LightGBM is easily interpretable via SHAP: you see which features drive failure risk (temperature, vibration, reboot count).

How Federated Learning Protects Data

For medical devices with HIPAA/GDPR requirements, raw data cannot be transmitted. Federated learning trains local models on each device, sending only weight updates to a central server. This eliminates private data leakage.

Work Process: From Audit to Monitoring

What's Included in Deliverables

• Architecture and pipeline documentation.
• Trained model in ONNX or TorchScript format.
• Grafana dashboard: risk scores, inference history, alerting.
• Repository with code and configs for reproducibility.
• OTA model update instructions for devices.
• Two-day workshop for your team.
• 6-month warranty on production model (support and refinement).

Timelines and Cost

Timelines range from 4 weeks (basic classifier) to 4 months (full LSTM + dispatch + OTA). Cost is calculated individually—depends on number of devices, types, and required accuracy. Get a consultation—we will send an estimate and timeline for your project. Maintenance savings can amount to significant reductions per year per thousand devices. Order a pilot project—we'll deploy an MVP on your fleet in 2 weeks.

Why Choose Us?

6-month warranty on production model. MLOps certifications (Kubeflow, MLflow). Experience with LSTM and LightGBM in production. Average reduction of unplanned downtime: 30%.

When does a time series forecasting model fail in production?

The CFO requests a quarterly sales forecast. An analyst builds SARIMA on three years of data, achieves MAPE 8.3% on the test set, and deploys. Two months later, the metric in production jumps to 23%. The root cause: the model was trained on pre‑COVID data, tested on a stable period, but production hit a promotion and supply chain disruption. Data leakage plus distribution shift—perfect notebook numbers, a broken forecast in reality. We have seen this pattern dozens of times across retail, fintech, and IoT. Our team has delivered more than 50 forecasting projects over 5+ years.

Incorrect cross-validation. Standard train_test_split for time series creates data leakage: the model sees future values during training. The correct approach is TimeSeriesSplit or walk‑forward validation with an expanding window.

Multiple seasonality. Hourly electricity consumption has three seasonalities: daily (24h), weekly (168h), yearly (8760h). SARIMA handles only one. Prophet can handle multiple but scales poorly to thousands of series.

Missing values and anomalies. A missing sensor reading is information (the sensor turned off), not NaN. Linear interpolation destroys this signal. Proper handling depends on the missingness mechanism.

Cold start. A new SKU in a 50,000‑item assortment has no history, yet a forecast is needed. Standard approaches fail; cross‑learning or feature‑based methods are required.

Why is model selection critical for your data?

Prophet (Meta) – a solid start for business data with clear seasonality and holidays. Fast setup, interpretable, built‑in outlier detection. Fails on irregular patterns and does not scale beyond ~10k series without parallelization.

Gradient boosting on features (LightGBM, XGBoost) – often underestimated. Engineer lags (t‑1, t‑7, t‑28), rolling means, day‑of‑week, holidays. The model trains on all series simultaneously, solving cold start via transfer learning. MAPE in retail often beats neural nets with proper feature engineering.

TFT (Temporal Fusion Transformer) – a transformer designed for interpretable forecasting with covariates. Built‑in variable selection, temporal attention, quantile outputs. Available in pytorch‑forecasting. Requires ~10,000+ records per series for stable training.

PatchTST – splits the series into patches (like ViT for images), capturing local patterns better than classic transformers. Excellent for long‑horizon forecasting (96–720 steps ahead).

N‑HiTS, N‑BEATS – attention‑free neural architectures, faster than TFT, competitive accuracy. N‑BEATS won the M4/M5 benchmarks for tasks without covariates.

How do we deploy TFT in production?

A typical pipeline via pytorch‑forecasting:

training = TimeSeriesDataSet(
    data,
    time_idx="time_idx",
    target="sales",
    group_ids=["store", "sku"],
    min_encoder_length=max_encoder_length // 2,
    max_encoder_length=max_encoder_length,  # 120 days
    min_prediction_length=1,
    max_prediction_length=max_prediction_length,  # 28 days
    static_categoricals=["store_type", "category"],
    time_varying_known_reals=["price", "promo_flag"],
    time_varying_unknown_reals=["sales"],
    target_normalizer=GroupNormalizer(groups=["store", "sku"], transformation="softplus"),
)

A common mistake: the default target_normalizer (StandardScaler) breaks predictions for series with zero values (no sales on weekends). GroupNormalizer with transformation="softplus" is the correct choice for count data.

Case study: retail demand forecasting

A chain of 120 stores, 8,000 SKUs, 28‑day forecast horizon. The original system: SARIMA per series, MAPE 18.4%, retraining cycle – 6 hours. We replaced it with TFT on PyTorch + pytorch‑forecasting: a single model for all series, MAPE 11.2%, retraining – 40 minutes on an A10G. Feature importance via variable selection revealed that day_before_holiday influences more than the holiday date itself. Annual savings on inference alone exceeded $50,000.

Step‑by‑step configuration

1. Data collection and preparation. Handle missing values (mark NaN, interpolate only for technical failures), aggregate to required frequency, engineer covariates (holidays, promotions, prices).
2. Create TimeSeriesDataSet. Set group_ids (store + SKU), time index, forecast horizon. Choose target_normalizer based on target distribution.
3. Train a baseline. Prophet or LightGBM first – to understand complexity.
4. Train TFT. Use TemporalFusionTransformer with loss=QuantileLoss(), tune learning rate and hidden layer sizes.
5. Validate and interpret. Walk‑forward test, analyze variable selection, build attention heatmaps.

How to properly evaluate forecast quality?

RMSE alone is misleading – it over‑penalizes large values. Our standard set:

• MAPE – interpretable, unstable near zero.
• sMAPE – symmetric, avoids division by small numbers.
• MASE (Mean Absolute Scaled Error) – normalized relative to a naive seasonal forecast, ideal for comparing series of different scales.
• Pinball loss – for probabilistic forecasting, inventory management.

We guarantee a model card with these metrics on the validation set and walk‑forward results on at least 6 months of history.

What deliverables do you receive?

• Documentation of chosen architecture and hyperparameter rationale.
• Reproducible training and inference pipeline (Docker + CI/CD + Airflow/Prefect).
• Committed code with unit tests for key components.
• Team training: retraining, output interpretation, deployment of new versions.
• 3 months of post‑delivery support (consultations, bug fixes, fine‑tuning).

The model is deployed via FastAPI or Triton Inference Server. Retraining is scheduled (e.g., weekly) via Airflow with drift validation and automatic rollback if metrics deteriorate.

Process and timeline

We start with EDA: visualization, ADF test, STL decomposition, analysis of missing values and outliers. This takes 2–3 days but often reveals systemic data issues that block forecasting. Then we build a baseline (naive seasonal, Prophet), engineer features for LightGBM, and select a neural architecture if needed. Walk‑forward validation with a realistic horizon. Deployment via API with automatic retraining scheduled via Airflow or Prefect.

Timeline: MVP forecast on one data type – 3–6 weeks. Hierarchical forecasting system with automation – 2–5 months. Cost is calculated individually based on data volume, number of series, and required accuracy.

Our team consists of certified ML engineers (AWS ML Specialty, GCP Professional ML Engineer) with 5+ years on the market and over 50 completed forecasting projects. Contact us for a free analysis of your data – we will assess the task and provide initial recommendations within 1–2 days. Request a consultation to ensure your forecasts work in production, not just in a notebook.