JCBTech / Products
[ Products ]

Products.

Industrial intelligence. Deployed locally. No cloud required.

[ 01 ] · Products

Five products. One philosophy.

Every product runs entirely on your hardware. No cloud subscriptions, no data leaving your network, no internet dependency. From document retrieval to real-time process monitoring, machine learning anomaly detection, ceiling-mounted safety scanning, and OT cybersecurity assessment — each solution is designed to deploy once and run indefinitely.

llm · rag · document retrieval
[ 02 ] · AI Maintenance Assistant

AI Maintenance Assistant

A locally hosted AI that retrieves accurate, sourced answers from your technical documentation. Upload equipment manuals, wiring diagrams, and maintenance procedures — then ask questions in plain English. The system finds the relevant section, cites the source, and delivers the answer without ever connecting to the internet.

  • Local LLM inference. Runs a quantised Phi-3 language model directly on the host machine. No API keys, no cloud, no data exposure. Responses in 15–60 seconds depending on hardware.
  • RAG knowledge base. Documents are embedded into a vector database using QA-optimised transformers. The system retrieves the most relevant passages before generating a response, ensuring answers are grounded in your actual documentation.
  • Multi-format support. Processes PDF, DOCX, XLSX, and TXT files. Integrated OCR handles scanned documents and legacy paperwork automatically.
  • USB deployment. Packaged as a standalone Windows executable. Copy to a USB stick, plug into any machine, and run. No installation, no admin rights, no Python required on the target.
  • Sector-agnostic. Equipment manuals, compliance documents, diagnostic procedures, technical standards — any industry where personnel need fast, accurate access to documentation without network connectivity.
Visit JCBTech Assistant
scada · ml · llm · industrial
[ 03 ] · PLC Monitor

PLC Monitor

A full industrial SCADA platform with integrated machine learning and AI analysis. Connects directly to Siemens S7, Modbus TCP, and OPC UA PLCs — polling live process data, managing alarms, charting trends, and running anomaly detection in real time. The built-in SCADA editor lets you design custom HMI screens without writing code.

  • Live process monitoring. Real-time dashboard displaying process values, digital status, alarm banners, and KPI tiles. Data logged to SQLite with configurable write intervals, age-based pruning, and automatic archiving.
  • SCADA editor. Drag-and-drop screen builder with gauges, bars, tanks, motors, valves, pumps, and indicators. Bind any widget to a PLC tag. Dynamic alarm borders, pulsing glow effects, group frames, and custom colour pickers. Publish screens accessible from any browser.
  • Machine learning. Dual-detector anomaly detection using Isolation Forest and Local Outlier Factor. Named tag groups with independent models and sensitivity. Trend prediction with quality filtering. Correlation shift detection. Health score history charted over time. Per-anomaly detector identification shows exactly which algorithm flagged each event.
  • AI analysis. Integrated LLM generates plain-English process assessments, shift handover summaries, and answers operator questions using live data, ML results, and RAG documentation as context. Hallucination prevention built in.
  • Multi-protocol. Siemens S7 via Snap7, Modbus TCP, OPC UA, and a built-in simulator for testing. Switch protocols by changing one line in the configuration file.
ml · esp32 · retrofit · lightweight
[ 04 ] · ML Monitor

ML Monitor

A lightweight anomaly detection appliance designed for retrofit installation on existing equipment. No LLM, no SCADA editor — just machine learning, process data, and a single-page dashboard. Connects to any Modbus TCP device or pairs with our ESP32 smart sensor firmware for a complete monitoring solution under thirty pounds per node.

  • Dual-detector ML. Isolation Forest detects global outliers — values far from all training data. Local Outlier Factor catches contextual anomalies — unusual value combinations even when individual readings appear normal. Health score takes the worst of both detectors, with per-anomaly identification showing which algorithm triggered.
  • ESP32 smart sensor. Custom firmware turns an ESP32-S3 into a WiFi-connected Modbus TCP IO module with local intelligence. Sixteen-times oversampling, moving average smoothing, rate-of-change detection, RMS vibration calculation, crest factor analysis, zero-crossing frequency counting, min/max tracking, signal validation, threshold alarms, rate alarms, and alarm latching — all running locally in microseconds without network dependency.
  • Non-intrusive retrofit. Clamp-on current transformers, surface-mount temperature probes, bolt-on vibration sensors, and optocoupler-isolated digital inputs. Monitor existing equipment without touching the PLC wiring or modifying the control system.
  • Trend and scatter visualisation. Overlay up to four signals on a single trend chart with independent Y-axes and baseline bands. Interactive scatter plot maps two-variable relationships with colour-coded anomaly classification and P5–P95 baseline zones. Both refresh in real time.
  • Deploy anywhere. Runs on a Raspberry Pi, packages as a standalone Windows executable, or builds as a locked deployment with embedded configuration. CSV export includes readings, ML health history, detector status, anomaly events, baseline statistics, predictions, and correlations.
lidar · zone · fail-safe · plc interlock
[ 05 ] · LiDAR Zone Monitor

LiDAR Zone Monitor

A portable safety detection system using ceiling-mounted 3D LiDAR for automated zone monitoring and PLC interlock. No vision processing, no neural networks — just trigonometry, point cloud clustering, and deterministic fail-safe outputs. Mounts at ceiling height, detects objects by ground distance, and writes zone states directly to Siemens S7 or Modbus TCP PLCs.

  • Fail-safe output design. Every PLC bit is HIGH when safe. Object detection, sensor fault, comms loss, or processor crash all drive outputs LOW. The PLC never needs to distinguish failure modes — any LOW means stop. System life bit, per-sensor health watchdog, zone-clear confirmation, and individual zone bits all follow the same convention.
  • Ceiling-mount geometry. Sensor mounted at five to six metres, vertical scan plane sweeping from straight down toward horizontal. Ground distance calculated as height times tangent of scan angle. Three configurable zones — near, mid, far — defined by ground distance with independent boundaries per sensor. Object size filtering with configurable minimum threshold.
  • Multi-sensor coverage. Up to four independent ethernet LiDAR sensors, each with its own mount height, field of view, detection width, and zone boundaries. Dashboard shows dual view per sensor: vertical cross-section and top-down floor plan with real-time point rendering and object tracking.
  • Comms health watchdog. Per-sensor health based on receiving valid scan data within a three-second timeout. No data means no health signal — the PLC sees the fault immediately. Supports Ouster, SICK, Livox, and Hokuyo hardware with a pluggable driver architecture.
  • Deploy anywhere. Open build for site setup with live config editing and save-to-file from the browser dashboard. Locked build embeds the finalised configuration into a standalone Windows executable — no Python, no config files, no source code exposed. Runs from USB on any Windows PC.
nist csf 2.0 · ot-aware · conformance
[ 06 ] · JCBTech CSF

JCBTech CSF

A guided NIST Cybersecurity Framework (CSF) 2.0 assessment tool, purpose-built for OT environments. It ingests documentation and third-party tool output, performs low-risk passive testing, maps findings to the official CSF 2.0 core, and produces a human-signed-off report with prioritised, OT-aware hardening recommendations. Runs entirely offline — nothing leaves the network.

  • Guided four-step workflow. Intake, scan, gaps, report. It records scope and sign-off, tests what it safely can, then interviews the assessor about everything it couldn't detect — so a purely interview-led assessment is never a dead end. Passive testing meaningfully covers Identify, Protect, and Detect; the questionnaire fills Govern, Respond, and Recover.
  • Passive, OT-safe collection. Parses Nmap, Nessus, and OpenVAS output, reviews exported switch, firewall, and PLC configurations, and observes the network without ever injecting traffic into the OT VLAN. Every finding maps deterministically to a CSF subcategory validated against the official core — the model never invents a finding.
  • OT-aware recommendations. For OT assets the advice inverts IT defaults: contain and monitor rather than patch-now, compensating controls over unplanned downtime. Hardening guidance is drawn from NIST SP 800-82 and IEC 62443, not generic IT baselines against plant that cannot take an outage.
  • Multi-framework cross-references. Every gap is mapped to the controls that remediate it in NIST SP 800-53, ISO 27001:2022, and IEC 62443 — one CSF assessment, three framework views, and a genuine UK regulatory edge. "Not assessed" is a first-class result that becomes the client's data-collection to-do list.
  • Signed, offline deployment. Runs from USB on Windows, Linux, and Raspberry Pi with no cloud dependency at runtime. Ed25519-signed licensing, an encrypted-at-rest findings store, and a self-contained HTML report. Human sign-off is required on every finding — advisory by design, never auto-certified.