CNC monitoring has traditionally required MTConnect or OPC-UA protocol integration to pull spindle data, feed rates, and alarm codes from machine controllers. That integration model works well in shops with standardised, modern CNC equipment from a single vendor. It breaks down in job shops and automotive component suppliers running 15-year-old Fanuc, Siemens, and Mitsubishi controllers on the same floor, where each controller requires a separate integration and a protocol translator that adds cost and failure points.
CNC machine monitoring software built on computer vision sidesteps the protocol dependency entirely by extracting machine state, cycle counts, and operator activity from camera observations rather than controller connections.
What does CNC machine monitoring need to track?
Effective CNC monitoring captures four categories of operational data:
Spindle state: running, stopped, air cutting, in tool change, in setup, or in alarm condition. This drives the Availability calculation in OEE. Most CNC machines spend 30-40% of available time in non-spindle states according to AMT data; monitoring makes those gaps visible.
Cycle count and cycle time: actual cycles completed per shift versus scheduled, and actual cycle time versus programmed cycle time. Cycle time deviation is the primary Performance loss signal in CNC operations. A 12-second cycle running at 14 seconds is losing 16% of Performance and generating no alarm.
Tool change frequency and duration: tool changes that take longer than the programmed duration indicate worn tools, incorrect tool length, or setup errors. In high-volume machining, tool change overruns generate significant takt loss.
Operator activity at the machine: when the machine is stopped, is an operator present? Is setup being performed or is the machine unattended? Machine stoppages without operator presence are a distinct loss category from stoppages where the operator is actively working.
How camera-based CNC monitoring works without PLC access
A camera positioned to observe the CNC machining area captures two types of visual signal that allow state inference without controller connection:
Coolant flow. Active machining generates coolant flow that is visually distinct from the dry state of a stopped or idle machine. In most machining centres, coolant flow is directly correlated with spindle engagement. A camera that detects coolant presence is effectively detecting spindle state with 90-95% reliability in standard machining environments.
Chip generation and workpiece movement. Active machining generates chips and moves the workpiece or table. For turning operations, a camera observing the chuck area can detect rotation and coolant flow as proxies for spindle state. For milling, table movement and coolant spray are the primary visual signals.
Operator and door presence. Machine door state (open for loading/unloading, closed for machining), operator position at the machine, and whether an operator is performing setup or standing idle are all camera-observable events that inform state classification.
Combined, these visual signals allow a camera-based system to achieve 92-96% accuracy in classifying CNC machine state without any controller connection, based on Jidoka Tech’s deployment data.
What camera-based monitoring does not replace
Camera-based CNC monitoring is a production monitoring tool, not a machine health monitoring tool. It answers the question “is this machine producing, and at what rate?” It does not answer the question “is this spindle showing early signs of bearing wear?”
For predictive maintenance based on vibration signatures, spindle current draw, or thermal imaging, dedicated sensor-based tools remain necessary. Camera-based monitoring and vibration-based predictive maintenance are complementary, not substitutes.
Similarly, for shops where cycle time variance needs to be traced to specific NC program line numbers, spindle load data, or feed rate variations, MTConnect integration with the controller remains the right approach, provided the controller supports it.
Deployment considerations for mixed-vintage CNC shops
The practical advantage of camera-based monitoring appears most clearly in shops with mixed controller generations. A job shop running 10 machines across three controller vendors and four vintages can deploy camera-based monitoring across all 10 machines in 2-3 weeks. An MTConnect integration covering the same 10 machines, if all 10 even support the protocol, typically takes 2-4 months and generates ongoing maintenance requirements as controller firmware updates break integrations.
For a shop where the monitoring goal is OEE visibility and spindle utilisation tracking rather than deep process analytics, camera-based monitoring covers 85-90% of the value at 30-40% of the implementation cost and timeline.