Machine monitoring in manufacturing has a gap nobody talks about on vendor demo calls. Your machine monitoring system reports 76% OEE. The spindle ran all shift. Cycle times looked normal. Then your supervisor finds a torque step that got skipped at station 3. Sixty parts in, rework pile growing.
The global machine monitoring solutions market sits at $3.78 billion in 2025 . Most of that spend captures what machines report. Not what actually happens on the floor.
Machine monitoring is the automated, continuous capture of machine state, cycle time, downtime, OEE, and condition signals, displayed live to operators and supervisors. It replaces end-of-shift clipboard reconciliation with data you can act on during the same production window.
This guide explains the three connection methods, the eight metrics worth tracking, and where real time machine monitoring stops and process visibility begins.
What Machine Monitoring Measures and The Layer it Cannot See
Machine monitoring gives you a live feed of your machines. Run state, cycle count, downtime duration, OEE broken into its three components. That data is accurate. The problem is its scope.
A machine monitoring system reports machine behavior, not human behavior. Those are two different things, and confusing them is where most monitoring deployments stall.
What it sees:
- Run/stop state and availability against scheduled time
- Performance rate vs. programmed nominal speed
- Equipment condition monitoring signals: vibration, temperature, current draw
- Unplanned stop events, logged with timestamps
What it cannot see:
- Whether the operator completed the correct assembly sequence
- Whether the right component was installed
- Whether a process step was skipped entirely
Over 64% of industrial facilities now use machine monitoring systems, yet nearly half still lack process-level visibility.
Machine sensors report what ran. They don't report what was done right. That distinction is what the monitoring stack in the next section is built around.
The Monitoring Stack: Three Layers of Manufacturing Visibility
A single machine monitoring system won't give you production truth. It gives you Layer 1. There are two more layers above it, and the losses hiding in those layers don't show up on any OEE dashboard.
Layer 1: Machine monitoring
This is what most machine monitoring systems deliver out of the box. The machine reports availability, OEE, cycle count, downtime duration, and performance rate against programmed nominal speed. Data comes from PLC signals, IoT sensors, or protocols like MTConnect, OPC-UA, and Fanuc Focas.
What it misses: anything that doesn't register as a machine event. A CNC running at 85% of rated feed rate looks fine. A station skipping a torque step looks fine too. Layer 1 records neither.
Layer 2: Condition monitoring
This is the physical health layer. Vibration accelerometers, bearing temperature sensors, current clamps, and acoustic emission detectors sit on the machine housing without touching the PLC. They catch bearing fault frequencies, thermal gradients, and acoustic signatures that precede failure by 48 to 96 hours.
Vibration monitoring accounts for over 75% of industrial condition monitoring applications globally. Layer 2 tells you the machine is degrading. It still cannot tell you what the operator did at station 3.
Layer 3: Process and operator monitoring
This is the gap layer. AI-enabled cameras analyse operator actions frame by frame against a digital SOP, confirming component presence, assembly sequence, and step timing in real time. Statistical studies across automotive, electronics, and food processing consistently show that human inspectors catch between 70% and 85% of defects under optimal conditions, with that figure dropping sharply during night shifts and end-of-shift periods.
Layer 3 is where rework originates, and it is the layer no machine signal can cover.
The right method for connecting your machines depends entirely on which layer you are targeting and that is what the next section covers.
Three Methods For Connecting Machines: Protocol, Sensor, and Vision
Your fleet is mixed. Some machines have Fanuc controllers. Some are 1990s lathes with no digital output. Some are assembly stations where the operator is the variable. Every machine monitoring deployment starts with the same question: which connection method fits which asset? Pick the wrong one and you get blind spots, wasted budget, or both.
Method 1: Protocol-based (MTConnect, OPC-UA, Fanuc Focas, Siemens SINUMERIK)
This is the richest form of cnc machine monitoring available. The machine monitoring system reads directly from the controller: spindle load, axis utilisation, M-code status, feed override, alarm codes, actual vs. programmed cycle time. Data comes from the control itself, not an inference from external signals.
Limitation: only works on machines with compatible controllers. Older equipment stays dark.
Method 2: IoT sensor-based (current clamps, vibration, optical)
Equipment condition monitoring via external sensors requires no PLC modification and no production stop. Current clamps and vibration detectors clamp onto the machine housing and infer run/stop state from electrical and mechanical signals. Works on any machine of any vintage: manual mills, band saws, injection moulders, 30-year-old lathes.
IoT-enabled predictive maintenance reduces breakdowns by up to 70% and cuts maintenance costs by 25%. Typical deployment for real time machine monitoring across a plant runs 48 hours with no production interruption.
Limitation: cannot read G-code, part programs, or native quality data.
Method 3: Vision-based (Edge AI + Production Line Inspection Camera)
A production line inspection camera paired with edge AI covers Layer 3 of the monitoring stack. It captures operator actions, component placement, assembly sequence, and dimensional checks frame by frame. These are the process failures that no machine signal can see.
Limitation: requires camera positioning and AI model training per station. Not a replacement for spindle-level equipment condition monitoring.
The Right Approach for Mixed Fleets
Protocol integration for CNC machines that support it. IoT sensors for legacy equipment. A production line inspection camera at process-critical workstations. That combination delivers complete machine monitoring coverage across the entire floor without forcing one method onto every asset.
Connection method sorted. Next: the eight metrics that tell you whether any of it is actually working.
The 8 Machine Monitoring Metrics That Drive Improvement Decisions
Every vendor demo leads with OEE. That number matters, but a single percentage score won't tell a plant manager where to intervene. These eight metrics are what experienced engineers actually track inside a machine monitoring system — ranked by operational impact, not alphabetical order.
1. OEE (Overall Equipment Effectiveness)
The master metric. Availability × Performance × Quality. The global average across discrete manufacturing sits at approximately 68% in 2025, up from 58% in 2020. Only 6% of manufacturers sustain 85% or above. A 10-point OEE improvement on a $50M production plant recovers roughly $5.5M in throughput without capital expenditure.
2. Availability rate
Scheduled time vs. actual run time. A low score points directly to unplanned downtime or excessive changeover. This is the first number to investigate when real time machine monitoring dashboards show OEE dropping mid-shift.
3. Performance rate
Actual cycle time vs. programmed nominal speed. Manual OEE tracking overstates performance by 8 to 12 percentage points compared to automated measurement. A cnc machine monitoring system catches a machine running at 85% of rated feed rate. A clipboard round never will.
4. Quality rate
Good parts vs. total output, logged at machine level. Aggregating scrap at end of shift masks which machine and which operator produced the problem.
5. Changeover time
Setup starts to first confirm the good part. SMED-focused plants track this separately inside their machine monitoring system because it is a scheduled loss, not a breakdown. Hiding it inside availability distorts both numbers.
6. MTBF (Mean Time Between Failures)
A declining MTBF trend on a specific asset is the earliest signal that equipment condition monitoring needs to trigger a condition assessment. Waiting for the breakdown costs significantly more than acting on the trend.
7. MTTR (Mean Time To Repair)
Automated event detection reduces MTTR by 30 to 50% on detected failure modes, with payback for edge hardware typically within 2 to 6 months. Long MTTR almost always points to parts availability or technician scheduling gaps, not machine complexity.
8. Production vs. plan
Actual output against the production schedule. The most actionable metric for a shift supervisor. Machine monitoring makes this visible in real time rather than at the end-of-shift debrief when nothing can be recovered.
These eight metrics tell you where the losses are. Next we will cover what cnc machine monitoring unlocks specifically for machining centres, where the data goes considerably deeper.
CNC Machine Monitoring: Specific Setup Considerations For Machining Centres
Generic machine monitoring covers run/stop state and OEE. CNC machine monitoring goes deeper. A machining centre with a compatible controller hands you data that no external sensor can replicate. Here is what that looks like in practice and where the setup decisions actually sit.
1. Protocol data available from the controller
CNC machine monitoring via Fanuc Focas, MTConnect, or Siemens SINUMERIK exposes spindle load, axis utilisation, feed override setting, M-code status, alarm codes, and actual vs. programmed cycle time. This is the controller speaking directly. No inference, no estimation.
2. Spindle health via condition monitoring
Vibration accelerometers on the spindle housing detect bearing fault frequencies and thermal gradients. AI-driven predictive maintenance delivers a 70 to 75% decrease in unplanned downtime and a 10 to 20% improvement in OEE on CNC cutting operations. Most equipment condition monitoring platforms predict spindle failure 48 to 96 hours in advance with accuracy above 85%.
3. The micro-stoppage problem
This is where most cnc machine monitoring deployments find their first surprise. Four to seven second stops for chip evacuation, tool changes, or bar-feeder resets are too short to log manually. They compound into 10 to 15% of available production time. Only real time machine monitoring captures them.
4. What the numbers look like before deployment
Typical OEE in small to mid-size CNC shops runs 30 to 60%. A 3 to 10 machine pilot typically improves OEE by 5 to 15% and reduces unplanned downtime by 20 to 50% within 3 to 6 months. Hardware and integration costs run $500 to $2,500 per machine, with SaaS fees of $50 to $300 per machine per month. Most shops report payback within 3 to 12 months.
CNC machine monitoring gives you the deepest machine-level data available. What it still cannot give you is visibility into what happens at the workstation after the spindle stops.
Where AI Camera Monitoring Closes The Gap Machine Data Cannot Close
Your machine monitoring system confirmed the spindle ran. Cycle time was nominal. OEE looked fine. Then the rework pile at final inspection told a different story. That is the Layer 3 gap: the distance between what the machine reported and what actually happened at the workstation. A production line inspection camera powered by edge AI is the only thing that closes it.
1. The process failure machine sensors cannot see
Machine monitoring confirms a machine ran at rated speed and produced output. It cannot confirm the operator installed the correct component, applied torque to the right fastener, or completed the assembly sequence in order.
Human inspectors catch between 70% and 85% of defects under optimal conditions, with that number dropping sharply during night shifts and end-of-shift periods. These are the escapes that generate rework, customer returns, and warranty claims.
2. What AI vision monitoring adds at Layer 3
A production line inspection camera with edge AI runs step-by-step verification against a digital SOP in real time. It confirms what was done and in what order, not just whether the machine cycled. That same frame-by-frame analysis catches process deviations the moment they happen, not at end-of-shift audit.
3. Nagare by Jidoka Technologies
Nagare converts existing RTSP and IP-compatible cameras (Bosch, Hikvision, and others) into a live real time machine monitoring system for process compliance. No new hardware required in most installations.
It runs two simultaneous streams on-premises via edge AI with no cloud dependency:
- Stream 1 (object detection): confirms component presence, correct bolt in correct position
- Stream 2 (action recognition): confirms operator movement, torque applied not just placed
Nagare delivers a 30% improvement in process adherence, 35% reduction in rework, and 25% drop in downtime. It tracks 100% of assembly steps through existing camera infrastructure. Skeleton tracking monitors operator actions without facial recognition, keeping operator identities private.
A complete machine monitoring stack covering all three layers gives you production truth. Layers 1 and 2 give you accurate OEE. Layer 3, covered by a production line inspection camera running edge AI, tells you whether the work was done right.
Conclusion
Most manufacturers running a machine monitoring system today have Layer 1 covered. They have OEE dashboards, downtime logs, and cycle time reports. The data is accurate. The problem is the floor does not stop generating losses at the boundary of what machine sensors can see.
Skipped assembly steps, wrong components, out-of-sequence work: none of it registers on a machine monitoring dashboard. That rework lands at final inspection, or worse, at the customer.
Jidoka Technologies' Nagare connects your existing cameras to the one layer most machine monitoring deployments leave completely dark. The data was always there. Now it is actionable.
Let's connect with Jidoka and put your existing camera infrastructure to work
FAQ
1. What is machine monitoring in manufacturing?
Machine monitoring is the automated, continuous capture of machine state, cycle time, OEE, downtime events, and condition signals displayed live to production teams. It replaces end-of-shift clipboard reconciliation with data operators and supervisors that can act on within the same production window.
2. What is the difference between machine monitoring and process monitoring?
Machine monitoring captures what the machine reports: availability, performance, and OEE. Process monitoring captures what the operator does: assembly sequence, component presence, and step timing against a digital SOP. A complete visibility system needs both. One without the other leaves recoverable losses invisible.
3. How do I connect legacy machines with no PLC to a monitoring system?
Wireless IoT sensors — current clamps, vibration detectors, or optical sensors — attach externally to any machine without PLC modification. A typical real time machine monitoring deployment across a mixed-vintage plant runs 48 hours with no production interruption and works on equipment of any age.
4. What OEE should I target when implementing machine monitoring?
The widely cited world-class benchmark is 85%, defined by Seiichi Nakajima in 1984. Most discrete manufacturers starting with machine monitoring operate at 60 to 68%. A realistic first-year target is a 5 to 12 point improvement within 6 months of deployment.
5. Can existing security cameras be used for process monitoring?
Standard CCTV records footage. It does not analyse production events. Nagare by Jidoka Technologies converts existing RTSP and IP-compatible cameras into a live production line inspection camera system using edge AI on-premises, tracking 100% of assembly steps without new hardware in most installations.
