Catch production faults
before they halt the line.
Upload your OEE reports, production logs or quality control spreadsheets. ThresholdIQ's 9 ML methods automatically detect equipment anomalies, yield drops, shift-pattern deviations and correlated sensor failures — no thresholds to configure, no historian required.
7-day unlimited trial · No credit card · Data never leaves your browser
Production teams spot faults too late
By the time a quality issue surfaces in your weekly report, the bad batch has already shipped, the equipment has already failed, and the downtime has already cost you.
Unplanned downtime from missed early signals
Bearing wear, pressure drift and temperature creep are all visible in your data days before failure — but only if something is actually looking. Manual inspection catches maybe 30% of early warning signs.
OEE drops buried in shift averages
A single shift's performance tanks due to a changeover issue or material defect. When averaged across three shifts, the weekly OEE looks fine — until the month-end roll-up shows a 4-point decline.
Quality escapes missed in SPC noise
Static ±3σ control limits don't account for tool wear drift or batch-to-batch material variation. Products ship out of spec while the chart shows "in control" — until a customer complaint arrives.
Correlated failures invisible to single-metric rules
Cycle time, reject rate and machine temperature all creep simultaneously — no single metric breaches its limit, but the process is deteriorating. Correlation monitoring catches this; static rules don't.
From production export to anomaly report in 60 seconds
No MES integration project. No historian query. No data scientist. Just upload the spreadsheet you already have.
Export your production or quality data
Pull your OEE report, SPC data, production log, or maintenance record from your MES, SCADA historian, or ERP system — or use the Excel file your floor supervisors already maintain. ThresholdIQ supports .xlsx, .csv, .json and .xml.
Upload — everything runs in your browser
Drag and drop the file into ThresholdIQ. All ML processing happens locally using browser-based Web Workers. Your production data never touches a cloud server — zero IT security risk, zero compliance exposure.
9 ML methods run automatically across every metric
Multi-window Z-score, EWMA spike detection, SARIMA seasonality, Isolation Forest, correlation deviation, DBSCAN clustering, trend detection, seasonal baselines, and stuck/zero sensor detection all evaluate every column simultaneously. No rules to write.
Review anomalies graded Warning, Critical or Emergency
Every flagged point is graded by severity with a signal breakdown showing which detection methods fired. Export as CSV for your maintenance CMMS, or generate a branded PDF report for your quality review meeting.
Works across the entire production floor
If your data has a timestamp and numeric metrics, ThresholdIQ finds what's abnormal — whether it's a sensor reading, a production count, or a quality measurement.
OEE Monitoring & Downtime Detection
Availability, Performance, and Quality components tracked per shift and per line. Trend detection flags gradual OEE decline before it becomes unplanned downtime. Isolation Forest catches multivariate failures — when throughput, cycle time and reject rate all deviate together.
Equipment & Sensor Fault Detection
Temperature, pressure, vibration, and flow rate anomalies detected automatically. Stuck-value detection catches sensor freeze. EWMA identifies sudden spikes. Trend detection flags bearing wear drift weeks before mechanical failure.
Quality Control & SPC Augmentation
Upload your quality inspection data — dimensions, weights, tensile strength, surface measurements. SARIMA seasonal residuals separate genuine out-of-spec events from tool wear drift. DBSCAN cluster noise identifies systematic defect patterns invisible to single-metric SPC charts.
Production Rate & Throughput Tracking
Units per hour, parts per cycle, batch completion times. Seasonal baselines account for planned shift changes and maintenance windows so routine stoppages don't fire false alerts. Correlation deviation flags when throughput drops while reject rate climbs simultaneously.
Energy & Utility Consumption
Electricity, compressed air, water, and gas consumption per machine or line. SARIMA separates production-shift seasonality from genuine energy anomalies. Correlated deviations flag when energy consumption rises without a corresponding increase in output.
Maintenance & MTBF Analysis
Work order counts, mean time between failures, preventive maintenance completion rates. Trend detection identifies increasing repair frequency on specific assets. EWMA catches sudden spikes in breakdown frequency that indicate accelerated wear.
What ThresholdIQ finds in a production report
This is the kind of multi-metric failure pattern that hides in plain sight on the shop floor — until ThresholdIQ flags it.
| Shift / Asset | OEE % | Reject Rate % | Cycle Time (s) | Spindle Temp °C | ThresholdIQ |
|---|---|---|---|---|---|
| Line A – Day | 87.2 | 1.4 | 42.1 | 68 | Normal |
| Line A – Night | 84.9 | 1.6 | 43.0 | 69 | Normal |
| Line B – Day | 79.1 | 4.8 | 51.3 | 81 | ⚠️ Warning — correlated deviation |
| Line B – Night | 71.3 | 8.2 | 58.7 | 94 | 🟠 Critical — multi-metric |
| Line C – Day | 88.5 | 1.2 | 41.8 | 67 | Normal |
| Line B – Morning | 58.4 | 16.7 | 0 | 0 | 🔴 Emergency — sensor failure + line halt |
ThresholdIQ detected early-stage thermal drift on Line B Night (Correlation Deviation + EWMA), escalated to Critical as OEE, reject rate, cycle time and temperature all deteriorated together, then fired Emergency when the spindle failed and the line halted — three escalating alerts that gave maintenance a 12-hour head start before total downtime.
Which ML methods fire on manufacturing data
All 9 methods run in parallel. Here's how each one surfaces different failure patterns specific to production environments.
Short, mid and long-term baselines evaluated simultaneously. Thermal runaway and pressure events that persist across all windows escalate to Emergency automatically.
Primary severity driverCatches sudden machine vibration spikes and instantaneous pressure anomalies that resolve quickly but are early indicators of mechanical wear.
Shock & spike detectionModels shift-based and maintenance-window seasonality. Ensures day-shift vs night-shift production differences don't trigger false alarms on night-crew throughput.
Shift-pattern awareDetects globally unusual combinations across all metrics — e.g. normal temperature + abnormal vibration + high reject rate together signals a balance issue invisible to single-metric rules.
Multivariate outlierMonitors relationships between OEE, temperature, cycle time and reject rate. When normally-correlated metrics diverge — high throughput but rising scrap — a fault is flagged immediately.
Process divergenceIdentifies process readings that don't belong to any normal operating cluster. Catches systematic batch defects, fixture misalignment patterns, and material lot outliers.
Systematic defect patternsMaintains separate baselines per shift (day/night/weekend) and per production run type. Tooling changeovers and planned maintenance windows are learned, not flagged.
Context-awareIdentifies monotonic drift in bearing temperature, tool wear metrics, and gradual OEE decline across consecutive measurement windows — days before the fault becomes visible to operators.
Early wear warningRepeated identical sensor readings or sudden drop-to-zero from a live metric signals sensor freeze, PLC communication failure, or complete equipment stoppage — fires Emergency immediately.
Sensor & line haltYour production data already knows
the fault is coming.
Upload your OEE report, quality log or sensor export. Nine ML methods find the anomaly pattern automatically — graded Warning, Critical or Emergency — before it becomes unplanned downtime.
Just enter your email — no password required
No MES integration needed · Works with any spreadsheet export · Data never leaves your browser