AI Vision Integration with PLC for Machine Control and Reject Mechanisms

Every AI vision system in the world can find a defect. The engineering problem in a production environment is getting the defective part physically off the line before it becomes finished goods. That requires the vision decision to reach the PLC, the PLC to write to the reject actuator, the actuator to complete its stroke, and the belt or conveyor to still be presenting the correct part to the reject station when the stroke finishes. Miss the timing budget by 20 milliseconds and the reject fires on the good part behind the defective one. Miss it by 100 milliseconds and the defective part ships. The difference between a vision project that works in production and one that lives in a pilot forever is almost always the millisecond budget.

The budget itself has four components that manufacturing engineers have to size carefully: capture latency (camera trigger to image ready), inference latency (image to pass/fail decision), transport latency (edge device to PLC over industrial protocol), and actuation latency (PLC output to actuator complete). Each component has its own physics. A pneumatic cylinder responds in tens of milliseconds. A servo-driven diverter responds in single-digit milliseconds. A high-speed air-blast reject is faster still. Choosing the right actuator for the line speed is as important as choosing the right camera or the right AI model.

The waterfall below shows a realistic timing budget for a discrete manufacturing line running at 60 parts per minute. Every bar is a component the controls engineer has to size. Add them up and the total decision-to-action window must land inside the physical spacing between the inspection point and the reject station on the conveyor.

Reading the waterfall from top to bottom tells a controls engineer where the trade-offs live. Camera capture is almost always fast enough — modern GigE Vision and USB3 Vision cameras deliver a full frame in under 10 milliseconds. AI inference on an edge GPU is next and is the single most tunable component: a heavier model produces higher accuracy but eats budget, while a distilled model runs faster but may need PLC-side verification to catch its false negatives. Transport over EtherNet/IP or PROFINET is deterministic and rarely a problem as long as the network is properly segmented. PLC scan and output latency are engineered around the specific controller family — a ControlLogix cycle is different from a CompactLogix cycle and different again from a SIMATIC S7-1500. The actuator stroke is the physics of the mechanism itself and cannot be shortened without changing the mechanism type. That is why the reject mechanism selection matters as much as the vision model choice.

The reject mechanism itself is the last link in the chain and the one that consumes the most time. Choosing the right mechanism for the line is a controls decision that depends on part weight, line speed, belt width, and whether the reject has to be sorted into multiple bins. The four common families below cover the vast majority of discrete manufacturing lines.

A common mistake plant teams make is over-specifying the mechanism. A high-speed air blast is exciting on paper but genuinely unnecessary on a line running at 40 parts per minute, where a simple pneumatic push cylinder does the job at a fraction of the maintenance burden. The opposite mistake is under-specifying: trying to make a pneumatic cylinder work on a 400 bottles-per-minute line always ends with the mechanism firing on the wrong bottle or missing rejects entirely. The right mechanism is the slowest one that reliably completes its stroke inside the timing budget for the line — that rule alone eliminates most of the reject-mechanism disasters seen in the field.

The protocol that moves the pass/fail decision from the vision system to the PLC determines whether the timing budget holds under load. Every PLC brand has a preferred fieldbus, and picking the wrong one adds middleware, latency, and failure modes. The three-column comparison below covers the protocols that account for the vast majority of vision-to-PLC integrations in 2026.

For plants that already own the PLC hardware, the protocol choice is usually made for them — Allen-Bradley plants use EtherNet/IP, Siemens plants use PROFINET, and mixed brownfield plants fall back to Modbus TCP or OPC UA. The more interesting decision is whether to add a secondary path for redundancy. On safety-critical or high-throughput lines, iFactory often runs the primary decision path over the native fieldbus and a parallel logging path over OPC UA into the historian, so a network hiccup on the primary path does not lose the audit trail. This dual-path pattern is invisible to the operator and adds no measurable latency, but it removes one of the most common causes of missed traceability data in field deployments.

A production-grade vision-to-PLC integration is not one signal but a handshake sequence. The PLC tells the vision system a part is approaching. The vision system captures, infers, and returns pass or fail. The PLC latches the decision, tracks the part through the belt shift register, and fires the reject at exactly the right moment. If any signal in the sequence is missed, the whole handshake breaks. The diagram below traces the six signals of a standard handshake.

The vision-to-PLC-to-reject chain looks the same on paper but differs sharply in physics from one industry to the next. The four use cases below are common iFactory deployments and show how the timing budget, mechanism, and protocol choice shift with the application. Each one is a real pattern iFactory has deployed across multiple lines.

The perspective below comes from a controls engineering lead who runs a mixed Rockwell and Siemens plant floor. It captures the operational payoff of getting the vision-to-PLC handshake right the first time.

The questions below are the ones controls engineers, automation leads, and plant IT teams ask most often when scoping a vision-to-PLC integration. Each answer is written for engineers who need enough technical detail to make a real decision rather than a marketing summary. If your specific PLC brand, line speed, or reject mechanism raises a question that is not covered here, the fastest path to a precise answer is a working session with the iFactory controls team.