Machine Learning Library

Five pure-JavaScript algorithms (linear/polynomial regression, logistic regression, Isolation Forest, KMeans, Bayesian Optimization) trained in the browser on OPC UA historian data. No Python, no cloud — one XML import.

Get notified at launch

Capabilities

Everything you need in one module

5 Algorithms in Pure JavaScript

Linear / polynomial (ridge) regression, logistic regression, Isolation Forest, KMeans and Bayesian Optimization — all running inside atvise (webMI + OPC UA + browser). No Python, no TensorFlow, no cloud, no third-party licenses.

Train Directly on Historian Data

Pick historized OPC UA signals as inputs and target. Automatic alignment and resampling (5 s to hours) with correct handling of slow-changing and binary signals — no export, no separate data-science project.

Regression & Prediction

Predict a continuous variable — e.g. outlet temperature as a function of power and flow — and review R², residuals and feature charts before putting the model to work.

Classification & Anomaly Detection

Logistic regression returns the probability of a binary state (e.g. failure in the next window); Isolation Forest flags anomalous vibration bursts without labeled data.

Operating-Mode Discovery (KMeans)

Cluster historian data to reveal production regimes and operating modes, with silhouette scoring and per-cluster inspection.

Bayesian Optimization of Setpoints

Search for the setpoints (power, flow, …) that maximize efficiency on top of a trained model — Optimize → Results workflow with iteration charts.

Portable JSON Models

Each model is a self-contained JSON (ml-portable-v2: pipeline + scalers + weights) that can be backed up and moved between servers.

Guided Workflow & Projects Tree

Data → Training → Performance → Inference, with metrics and charts at every step. Work is organized as groups → projects → algorithm instances; a simulation package demos the module without plant data.

Technical Details

Specifications & Requirements

Availability	Coming soon — pre-launch
Min. atvise® version	3.14
Algorithms	Linear/polynomial regression, logistic regression, Isolation Forest, KMeans, Bayesian Optimization
Runtime	Pure JavaScript in the browser — no Python, no cloud
Data source	OPC UA historian (historized signals)
Resampling	Automatic, 5 s – hours
Model format	Self-contained JSON (ml-portable-v2)
Deployment	Single XML import in atvise Builder
Browser	Chrome / Edge (modern)
External dependencies	None — 100% offline