Acceptance test plan

What others have used (references, not defaults)

• Get the required values from the controls owner before you test: the connection timeout, the loss it tolerates, and the longest gap it rides through. In a jointly published industrial design guide, standard I/O connections time out at 100 ms or more and tolerate up to three lost packets in a row.¹²³
• Confirm that the installed network matches the design, and that coverage meets the survey targets, before the acceptance run starts. A jointly published industrial design guide designed to at least -67 dBm RSSI and 25 dB SNR across the cell, and NIST notes that the team that installs a wireless network is often not the team that designed it.¹⁴⁵

What others have used (references, not defaults)

• Drive the paths the vehicles really take, through the places where roams happen: wall openings, metal tooling, storage, and machine aisles. NIST's factory propagation survey ran its receiver along routes that passed machines, circled storage areas, and ran behind walls, and Omron's AMR wireless guide lists transitional spaces and entries into metal tooling as predicted roam locations.⁶⁷
• Judge coverage from the vehicle's own radio and antenna along the whole path, not from a survey laptop. MiR states its signal minimums from the robot's perspective, -70 dBm to the best AP and -75 dBm to the second best, and Omron's guide reads weak signal on one AMR with strong signal in a survey tool as a reason to check that AMR's antennas and cabling.⁸⁷
• Run each route at the vehicle's real operating speed and record the speed beside every result. In a jointly published industrial design guide's attenuator-emulated roaming test, the maximum throughput a roaming client sustained fell from 15 Mbps at 5 mph to 4.5 Mbps at 100 mph.¹

What others have used (references, not defaults)

• Generate test traffic at the application's real interval and message size, not a generic ping. A jointly published industrial design guide set RPIs of 10 to 60 ms per test case to reach a target packet rate, and a university AMR study sent 64-byte pings every 50 ms to emulate a fleet manager or PLC flow.¹⁹²
• Load the channel with the other traffic it will carry in service, because the latency tail grows with load even when the average looks fine. In one university study, a roaming AMR client's 99.9th-percentile round-trip time rose from 116 ms on an idle network to 297 ms with 10 Mbit/s of background traffic each way per AP, and past one second when the APs shared a channel.⁹
• Test with the share of the fleet that will be active at peak, not one vehicle on an empty floor. 5G-ACIA's performance test method records the active device factor, the percentage of devices exchanging data at once, as a test condition, and Omron's AMR guide keeps a single-robot proof of concept separate from fleet acceptance testing.²⁷

What others have used (references, not defaults)

• Run long enough to catch the rare event, in days rather than minutes. A jointly published industrial design guide ran its tests continuously for one to two weeks to catch connection losses and maximum values, and NIST's deployment guide ties the run period to the application.¹⁴
• Count roams and samples as well as hours, because a 99.9th percentile means little without many samples behind it. One university AMR study ran 45 handovers per test and gathered more than 100,000 latency samples per configuration before it reported that percentile.⁹
• Treat a lab or quiet-weekend result as the best case, and run acceptance while the plant operates, with spectrum monitoring on. A jointly published industrial design guide says its own results came from a vacant 5 GHz channel with no outside interference and represent the best achievable performance, and NIST asks for validation with the existing traffic and while the plant is in operation.¹⁴

What others have used (references, not defaults)

• Measure loss and latency where the application sees them, between the controller or vehicle and its radio, not only in the AP's statistics. NIST recommends black-box tests that use live process signals and treats the application layer as where loss is perceived, and 5G-ACIA's test method places its measurement points at that same interface.⁴¹⁰²
• Log every roam with its position, AP, channel, signal, and round-trip time, and keep the controller's connection faults beside them. Omron's fleet tool records coordinates, base station and channel, signal, ping time, and BSSID changes for each map point, and a jointly published industrial design guide baselines RSSI, SNR, retries, CRC errors, channel utilization, and application timeouts.⁷¹
• Record the fleet manager's view of each vehicle's connection beside the network's view. Under VDA 5050, the broker publishes CONNECTION_BROKEN when a robot's connection ends unexpectedly, and each robot publishes its state on events or at least every 30 seconds.¹¹

What others have used (references, not defaults)

• Set pass lines on the maximum and a high percentile, never on the average. A jointly published industrial design guide notes that more wireless clients raise maximum latency while the average may barely change, and in its stationary tests 99.99 percent of samples stayed under 10 ms while single packets took up to 32 ms.¹
• One published pass line for a roaming client: total traffic convergence under 100 ms per roam, with no connection faults. A jointly published industrial design guide applied it with a 20 ms standard RPI and a 40 ms safety RPI and measured 63 to 97 ms in a best-case lab, so treat it as a reference point, not a target for your plant.¹
• Another published set of lines: no standard or safety connection faults over the whole run, no more than two packets lost in a row, and 95 percent of latency samples under half the RPI. A jointly published industrial design guide used these for its stationary wireless I/O tests.¹
• If CIP Safety crosses the link, a jointly published industrial design guide compared measured reaction time with calculated worst cases: the maximum below the single-fault value, and 99.99 percent of samples below the no-fault value. Those calculated limits come from the safety system's own tools and the machine builder's risk assessment, never from a wireless test.¹
• Robot makers publish their own network lines, and they differ widely. MiR caps round-trip time at 200 ms between robot and PC and 500 ms between robot and fleet server, with at most 2 percent packet loss, while Omron recommends 10 ms or less and allows at most 50 ms.⁸⁷