When considering CNC machine tending automation in a Thai factory, the first question is not “Which robot should I buy?” Instead, you must ask: Can the existing machine’s door be opened and closed automatically? Can the control system verify the status of the chuck or vise? Will the robot seat the workpiece without leaving chips behind? If you only compare payload and cycle time without clarifying these points, you’ll end up buying additional equipment to solve problems that stop your machines after installation. This article focuses on retrofitting robots to existing CNC machines, organizing the process from equipment survey, safety, RFP, FAT/SAT, to a 90-day proof of concept (PoC), in the order that factory managers should confirm.
What Does CNC Machine Tending Automation Actually Automate?
Tending involves supplying raw material in a specified orientation, loading it into the machine, removing the finished workpiece after machining, and transferring it to the next process or finished goods area. For lathes, the handoff point is the chuck; for machining centers, it’s the vise or fixture. The “automation scope” must include not just the robot’s transfer motion, but also door opening/closing, air blow, misplacement checks, chip removal, measurement, and recovery from errors—tasks that humans currently perform. If you automate only the robot’s transfer, the remaining tasks between the machine and operator will become bottlenecks later.
If you are considering new equipment and process design, see the separate article CNC Machine Tool Production Preparation for Thai Factories. This article focuses on adding robots to existing machines where machining conditions are already established. Also, we will not repeat the general comparison of robot types covered in Collaborative Robot Use Case Selection, but will focus on the interface with machine tools and acceptance testing.
FANUC lists loading/unloading, transfer to subsequent processes, and measurement/quality checks as typical machine tending tasks. This is useful for organizing the scope regardless of robot brand. However, what is possible in a FANUC case study may not be directly achievable on existing CNCs from other manufacturers. Haas’s September 2026 revision of its external robot integration documentation specifies required software, M-Fin, relays, auto doors, and safety settings for each model, and advises consulting an integrator for robot cell design. The first step is to confirm, for each target machine, the manufacturer, model, year of manufacture, control version, and list of options.
Initial Site Survey: Gather Data Before Deciding Robot Specs
Choose candidate machines not just because “operators are busy,” but because “waiting for loading/unloading actually limits capacity and the machine runs stably.” For at least two weeks, record—by timestamp—part number, lot, machining program, machining time, loading/unloading time, measurement, tool change, alarms, rework, and material waiting. If multiple part numbers are mixed, separate the data by part number. Average cycle time alone will not reveal long stoppages or setup variability. Also record day/night shift conditions, replenishment frequency, and setup authority to distinguish which stoppages are within the robot’s scope.
For example, if machining takes 20 minutes and manual loading/unloading takes 2 minutes, even if you reduce loading/unloading to zero, the theoretical maximum cycle reduction is only 2 minutes. In reality, robot door operation, grip confirmation, retreat, and detection delays further reduce this benefit. On the other hand, if operators patrol multiple machines and there is a long wait after machining ends, reducing “waiting time” is more effective than reducing the “2-minute task.” This is a hypothetical example for explanation, not an estimate of actual ROI. For evaluation, use the distribution of time from machining end to next cycle start, and stoppage time by reason.
Map the door width of the existing machine, distance to the loading port, interference with spindle/tool turret/fixtures, maintenance door swing, chip box replacement path, and material supply table replenishment path on a floor plan. Haas documentation recommends about 1 m of clearance around the machine for normal maintenance in its example layouts, but this is a recommendation in specific Haas documents, not a universal design value. Prioritize the actual maintenance manual, local safety assessment, and forklift/personnel aisles. If you squeeze in a robot by narrowing aisles but increase downtime for tool changes or chip removal, you defeat the purpose.
Connecting to Existing CNCs: Consider Seven States
A simple “We can communicate” is not enough for an order. The purchaser must create a table for these seven states: ① machining complete, ② spindle stopped/safe position, ③ door or window open/close, ④ chuck/vise open/close and grip, ⑤ robot entry/retreat, ⑥ cycle start permission, ⑦ emergency stop/safety door. For each, specify the signal sender, receiver, confirmation condition, and timeout handling. Control methods differ by machine: M-code and I/O, dedicated interfaces, industrial networks, etc. In any method, “command issued” and “state detected” are separate.
For example, after a door open command, do not allow the robot to enter until the open-end sensor is confirmed. After a chuck close command, confirm the condition for detecting workpiece grip. Haas documentation shows that sensor combinations differ between the chuck being fully closed empty and gripping material. This specific signal interpretation must be confirmed with manufacturer documentation and actual wiring diagrams, and should not be generalized to other machines. Also decide, after timeout, mid-cycle stop, power recovery, or emergency stop reset, which controller treats the current state as valid.
The normal automatic sequence is: machining complete → spindle stop/safe position → door open-end confirmation → robot entry → finished part grip → fixture release → finished part removal → material insertion → fixture close confirmation → robot retreat confirmation → door close-end confirmation → machining start. Write this out on paper. The actual order varies by machine and process. The key is to have “evidence to proceed” at each stage, and to stop safely on errors. Require the integrator to deliver a state transition diagram and I/O list. After handover, the site will have documentation to trace “why it isn’t running.”
Workpiece Supply and Gripping: Precision Is Not Determined by the Robot Alone
Gripper design starts with how dimensions, temperature, and surface change before and after machining. Raw material may have a rough surface; after machining, there may be oil film and burrs. Even with a two-finger gripper, decide the gripping reference, allowable grip force, and when to replace worn fingers. Whether you grip the inner diameter or outer diameter affects contact with machined surfaces and setup for the next process. Workpiece weight includes the material, residual coolant, and any gripping device in the fixture; also check the moment and posture at the robot wrist. Catalog payload alone does not determine suitability.
Whether you pick material from a bulk bin or supply it positioned in a tray affects cost and recovery ease. For small-lot, high-mix production where box stacking varies, consider vision recognition, but check samples for lighting, oil, reflection, overlap, and posture variation. Stable tray supply may allow you to avoid vision complexity. FANUC lists vision-based material positioning and post-removal measurement as use cases, but does not guarantee feasibility for individual parts. For the initial PoC, start with the most stable supply method and add complexity only as needed.
There is a difference between “workpiece inserted in fixture” and “seated correctly to the machining reference.” A single chip on the contact surface can cause dimensional errors. Combine workpiece presence detection, seating detection, chuck/vise pressure, clamp position, and touch probes as needed to define pre-machining conditions. Even if you add air blow, check the direction for chip escape, coolant splash, noise, and filter cleaning. If you exclude chip removal from the robot’s scope, unmanned operation time will be short.
Don’t Leave Doors, Chips, or Coolant as Equipment Boundaries
On existing machines, doors may be manual, openings may be small, and door open/close times may vary. Confirm with the machine tool manufacturer whether auto door addition is possible, and clarify warranty/modification scope, open/close-end detection, anti-pinch measures, and maintenance mode switching. Do not adopt proposals that bypass safety circuits for convenience. Never create a state where the machine mistakenly “thinks the door is closed” and starts machining. Haas’s M-codes and option numbers are examples for its NGC models and are not connection specs for other brands.
Chips affect workpiece gripping, sensor detection, floor slip, and fixture seating. The length of continuous operation may be limited not by the material tray but by the chip box and coolant tank capacity. Haas documentation also states that if the chip container is inside the cell, a discharge procedure is required. In the site survey, record chip shape, entanglement, discharge direction, box replacement frequency, coolant replenishment, and concentration checks. If sensors, cables, or grippers are exposed to splash, include covers, drainage, cleaning, and maintainability in the design.
Before extending unmanned operation time, check tool life, dimensional compensation, cutting fluid, material stock, and good/reject part storage. If you start night operation with tools near end-of-life, even high robot uptime will not maintain product quality. Whether you can preset spare tools or monitor wear depends on the machine tool’s functions and operation. For measurement, decide whether to simply display values, use them to stop the next process, or require approval for compensation. Include gauge calibration and master part checks in the operation.
Safety Design: Don’t Assume “No Fence Needed” Just Because It’s a Collaborative Robot
Hazards in the cell are not limited to robot movement. Rotating spindles, chuck opening/closing, dropped workpieces, sharp chips, coolant, compressed air, and unexpected restarts during maintenance are all risks. ISO 10218-2:2025 covers safety requirements for industrial robot applications and cells. The existence of a standard does not mean your specific cell is compliant. The purchaser and integrator must conduct a risk assessment covering the equipment, tasks, teaching, recovery, and cleaning, and confirm applicable laws, standards, and machine tool manufacturer conditions.
OSHA’s machine guarding documentation is a US reference and does not substitute for Thai law, but it is useful for checking points like contact with moving parts, flying debris, and intrusion into hazardous areas. Safety fences, interlocked doors, area sensors, etc., should be selected based on operator material replenishment, maintenance fixture adjustment, and chip box removal paths. Design so that “looks stopped” and “can be safely restarted” are distinct. If an emergency stop leaves a workpiece in the gripper or in front of the spindle, document who checks what before recovery.
Even with collaborative robots, machine hazards and sharp workpieces remain. Whether speed/force limits alone ensure safety must be evaluated including the process and end effector. In some cases, enclosing the cell is the most rational solution. In cells linking equipment from multiple vendors, clarify by contract who designs and tests emergency stop, safety door, and maintenance mode interlocks. Leave safety test results and residual risks as handover documentation.
Whether You Can Recover After a Stop Determines Actual Uptime
A good demo runs normal parts in sequence. But in reality, time is lost to exceptions: missing material, wrong orientation, double picking, grip failure, improper seating due to chips, out-of-tolerance measurements, tool end-of-life, etc. A cell that requires maintenance staff for every exception increases operational burden, even with fast cycles. The RFP should list not only normal operation but also frequent stoppage events and who is responsible for recovery. Alarm messages should specify not just “robot error” but the state, signal, and part number at which it stopped.
Recovery is designed in this order: remove the cause, check workpiece positions in the cell, synchronize machine and robot states, separate good/reject parts, and resume the cycle. To avoid double-processing in mid-process, you may need to manage part IDs, tray positions, and processed flags. Test that the system does not restart automatically after a power outage, and that after manual intervention, the fixture workpiece is re-detected. Provide recovery procedures in the local language with photos and state diagrams. For Thai factories, include Thai-language operation and error recovery procedures in the contract scope.
Uptime records should be created together with recovery design. Divide stoppages into “CNC machining alarm,” “robot grip failure,” “supply wait,” “measurement NG,” “safety door open,” “manual setup,” “planned stop,” etc., and record time and duration. This is to see not just robot uptime but whether the equipment produced the next good part. To clarify responsibility, synchronize log timestamps for machine, robot, and peripherals. In PoC analysis, treat long stoppages and frequent short stoppages separately.
Compare Cycle Time and ROI Using the Same Assumptions
For quotation comparison, use the machine’s “time from previous machining complete to next machining start” as the common metric, not just robot solo operation time. The breakdown is spindle stop/safe position, door open, finished part removal, material loading, grip confirmation, robot retreat, door close, machining restart. If you add air blow, probe measurement, compensation, or transfer to supply table, include them under the same conditions. Separate initial startup, changeover, tool change, and box replacement from the normal cycle, and multiply by frequency to reflect one-shift production capacity.
A simple evaluation formula is: (number of good parts × contribution margin) minus additional maintenance, consumables, electricity, quality loss, installation downtime, training, fixture changes, compared to the investment amount. Even if you can reassign staff to other processes, check whether capacity actually increases at the destination. The simple calculation “one operator eliminated” does not hold if replenishment, measurement, monitoring, and recovery remain. Only count the value of night operation extension for the actual measured time when quality and tool life are maintained. Do not quote payback years from general sources; use on-site data for sensitivity analysis.
For example, create a table with annual operating days, shift hours, order volume by part number, machining time, loading/unloading wait, good part rate, and maintenance cost as input fields, and calculate three cases: wait time halved, as expected, and zero. Include installation downtime and yield drop at startup to avoid only favorable numbers remaining. If the remaining service life of the existing machine is short, consider cell relocation/reuse potential. Even if multiple machines are proposed, stabilize signals and exception handling on one machine first, then recalculate flows and wait times for the second.
Documentation and Acceptance Criteria to Attach to the RFP
Don’t Assume “Just Change the Robot Program” for Part Changeover
In high-mix, low-volume factories, it’s more important to know who changes what, how long it takes, and what needs to be swapped for each part change, than how many part numbers can be handled. Treat tray pockets, gripper fingers, machine fixtures, machining programs, measurement programs, and finished goods dividers as a single recipe to prevent mix-ups. If only the machine program is switched but the gripper fingers remain from the previous part, it can cause dropped parts or misprocessing. If you use barcodes for part ID, decide what happens on read failure, manual entry authority, and how records are kept.
For the first target part, choose one with stable shape, clear grip and machining references, and steady demand. If you start with a difficult part with many exceptions, it’s hard to separate cell basic functions from part-specific issues. On the other hand, if you finish acceptance with only a sample of one part, you may overlook problems with tray capacity, finger change time, or measurement points for the next part. For the PoC, the second part should differ in dimension or orientation so you can verify changeover procedures. Include initial inspection after change, program verification, and prevention of material/finished goods mix-up in the test checklist.
Make Maintainability a Purchase Requirement
Maintenance cost estimates should include gripper wear parts, door drives, sensors, pneumatic devices, camera window cleaning, chip management, and backup creation. When the equipment stops and both the machine tool and robot alarms appear, ensure the primary troubleshooting order is clear. Confirm local spare parts inventory and lead time, remote support conditions, on-site response time, and night coverage before contract. Training should cover not just normal operation, but part changeover, material replenishment, quality issues, emergency recovery, and backup restoration.
For Thai factories with multilingual teams, it’s also important that short alarm names on screens, shop floor notices, and official maintenance procedures are consistent across languages. Simply handing over a translated manual may not match actual screen signal names. During SAT on-site training, have operators simulate errors and recover, and immediately correct any unclear screens or procedures. The boundary between “robot introduced” and “robot operational” is when results can be reproduced even if staff change.
To get comparable quotes from vendors, provide: ① CNC manufacturer/model/control version/options, ② part drawings and actual samples, ③ quantity/machining time/changeover frequency by part number, ④ current stoppage logs, ⑤ factory layout and transfer paths, ⑥ desired unmanned operation time, ⑦ quality judgment and measurement methods, ⑧ safety, maintenance, and training requirements. Photos alone are not enough to judge machine warranty conditions or signal availability. If manufacturer documentation is lacking, specify “requires actual machine confirmation” as a quotation assumption.
The comparison table should list not just “robot main unit price,” but also gripper, supply/discharge, door modification, machine-side modification, sensors, PLC/communication, safety fence, installation, software, measurement, FAT, SAT, training, spare parts, maintenance, and local support. Who will coordinate with the machine tool manufacturer is also important, as warranty and responsibility may change when modifying existing equipment. Before evaluating price, align the scope of excluded work and conditions for additional costs. Include in the contract who approves specification changes, change records, and delivery impact.
Acceptance criteria should be observable statements, not just “it runs.” For all specified part numbers, the system must perform defined grip, seating, and dimensional checks. On door open-end not detected, chuck grip not detected, safety door open, material missing, measurement NG, etc., the system must transition to the expected stop state. After removing faults, the system must be restartable by defined procedures. Target production capacity must be evaluated under specified part mix, material replenishment, tool change, and planned stops. Use the same test checklist for FAT and SAT. Agree on numerical targets only after establishing on-site baselines, not from vendor catalogs.
FAT, SAT, and 90-Day PoC as a Unified Verification Plan
FAT (Factory Acceptance Test) is pre-shipment acceptance at the vendor’s site; SAT (Site Acceptance Test) is post-installation acceptance at your site. In FAT, use representative parts and simulate not only normal cycles but also grip failure, material shortage, door not open, fixture unclamped, emergency stop, and power recovery. In SAT, use the actual CNC, coolant, chips, lighting, factory network, and material variation to check differences from FAT. Decide safety test responsibilities and methods, and keep records. Simply shooting a video for “operation check” is not enough for maintenance handover.
A practical 90-day PoC plan is: stabilize one part in day shift with supervision in the first 30 days, expand to more parts and exception conditions in the next 30 days, and confirm actual shift capacity including replenishment, maintenance, and recovery in the final 30 days. This is a recommended verification breakdown, not a universal period for all factories. Set gates for each stage, and do not extend unmanned time if quality, safety, or stoppage reasons remain unresolved. Weekly, record good part count, stoppage reasons, recovery time, and manual interventions, and document program and work standard adjustments.
For final acceptance, receive drawings, electrical/pneumatic diagrams, I/O lists, state transition diagrams, risk assessment, test checklists, robot and PLC backups, spare parts lists, training records, maintenance contacts, and recovery procedures in the local language. Do not leave settings known only to commissioning staff before expanding to new parts or equipment. Also create a single table showing where to report failures (machine tool maker, robot maker, integrator) and who does primary troubleshooting.
Operational Rules the Client Should Establish: Clarifying Responsibility During Downtime
If a project is closed simply because the integrator demonstrates normal operation at handover, responsibility for downtime will be pushed back and forth among the machine tool, robot, and on-site personnel every time a stoppage occurs. To prevent this, designate a primary checker and contact point for each signal. For example, “No machining completion signal” falls under the machine tool side, “Unable to pick up material” is the supply/gripper side, and “No door closed confirmation” is the responsibility of the door drive and sensor side. This table is not for assigning blame to one party, but for indicating what should be checked first. If the cause lies at a boundary, ensure that logs can be jointly reviewed by recording the time, part number, program number, alarm, and the signal immediately preceding the stoppage.
Authority to restore operations on-site should be divided by role. For example, can operators only replenish materials and perform predefined simple resets, or are they also allowed to remove workpieces from jigs? What tasks are restricted to maintenance personnel—inspecting safety doors, adjusting sensor positions, or reteaching the robot? For which actions is quality assurance approval required—reprocessing after a failed measurement, or releasing quarantined items? Without such boundaries, there is a risk that efforts to resume operations quickly may compromise safety conditions or traceability. Manuals should include not only button operations, but also the conditions under which stopped workpieces can be restored as good products, when they must be scrapped, and how to record carryover to the next lot.
When requesting quotations, make sure to specify what is *not* included. Costs for modifying existing CNCs by the manufacturer, lead times for automatic doors, on-site support at night, tool life management, coolant replenishment, network installation, inspection equipment calibration, and third-party safety evaluations may all be quoted separately depending on the proposal. If any of these are excluded, specify who will arrange them and by when in the project schedule. In particular, clarify in writing before ordering how manufacturer warranties will be handled in the event of machine tool modifications. Only by filling in these excluded items when comparing total estimates from multiple vendors can you make a valid comparison of implementation costs and lead times.
At inspection meetings 30, 60, and 90 days after acceptance, compare the top reasons for stoppages, yield rates, manual interventions, average recovery times, and maintenance delays using consistent definitions. Simply counting “the number of robot stoppages” does not evaluate equipment performance. If the robot is merely waiting while the machine tool is stopped, the number of robot failures may be low, but the yield will not increase. Conversely, it is also a mistake to treat the number of times abnormalities were quickly detected and safely stopped as a simple negative. Separating capability, quality, safety, and recoverability helps determine whether to roll out the system to other equipment.
If the target performance is not achieved, first categorize the causes as “equipment-specific issues,” “parts supply issues,” “machining condition issues,” or “operational procedure issues.” Then verify which countermeasures—design changes, supply method reviews, training, or process adjustments—are effective for each cause. Rather than hiding missed projections, it is more useful for future investment decisions to clarify causes and improvement costs with a small number of part numbers. Documenting these learnings is itself a deliverable of the 90-day PoC.
Aligning Data Definitions Before Using Measured Values
Before reporting in a PoC that “the number of good products has increased,” it is essential to standardize what is being counted as a single unit. The number of machining completion signals, the number of workpieces placed on a tray, the number of inspection-passed items, and the number of shippable items may not always match. Decisions must also be made regarding how to handle reworked items or those that underwent manual intervention during the process. In operations where a product is considered good when the robot places it on the finished goods tray, there is a risk of sending dimensionally defective or misidentified items to the next process. Nonconforming items identified by measurement should be physically isolated and cross-referenced with the part number, timestamp, and measured values in the log. In summary tables, list the target part numbers, shifts, input quantities, machining completions, inspection passes, defect counts, downtime, and manual interventions for the same period.
The reference date for performance evaluation should also be clearly specified. If the pre-implementation baseline is based only on specific busy days, and the post-PoC results are from small-lot production during a slow period, it is not possible to fairly compare equipment capabilities. If there are differences in material lots, machining programs, tools, measuring instruments, or work shifts, these differences should be recorded and reflected in the interpretation. Only cycles for comparable part numbers and conditions should be compared; the rest should be treated as separate validations. Both the client and the vendor should have access to the same raw data, and any changes to programs, gripper fingers, or sensor positions should be logged in the change history during weekly meetings. Even if the numbers improve immediately after changing a setting, unless the results can be reproduced in the next lot, they cannot serve as a basis for acceptance.
Furthermore, in continuous operation tests, do not consider the test passed solely based on “how many hours it ran without stopping.” Confirm whether the specified quantity was produced as good products, whether on-site personnel could restore operations according to procedures when stoppages occurred, and whether replenishment and chip removal were performed safely. If the integrator intervened manually during the test, record the status and reason at that time. This is to ensure that no hidden expert interventions remain between the acceptance test and routine operation. Use these results for the handover decision, and assign responsible persons and deadlines to any unresolved issues.
Frequently Asked Questions
Can robots be retrofitted to any existing CNC machine?
No. You must check door automation, safety circuits, I/O or communication, fixture/chuck state detection, maintenance space, and manufacturer modification conditions. Even if possible, it may not be cost-effective. Provide the machine model and control version to the machine tool maker and integrator and request an on-site survey.
Are safety fences or interlocks unnecessary if I use a collaborative robot?
The robot type alone does not decide this. Assess risks including the machine spindle, chuck, workpiece, sharp chips, and surrounding paths, and select necessary protective measures. You cannot assume safety just by limiting speed and force.
How do I estimate the cost and payback period for CNC machine tending?
Include not just the robot, but grippers, supply tables, door/fixture modifications, safety equipment, control integration, installation, testing, training, and maintenance. Calculate payback from actual measured values for loading/unloading wait, good part count, order volume, night operation, quality, and stoppage reasons for the target machine. General payback years not fixed to part mix and operating conditions are not useful for decision-making.
When can I start unmanned night operation?
After you can reliably reproduce both normal and error recovery in day shift, and have confirmed the capacity and management methods for material, tools, chips, coolant, and measurement. Also decide who receives alarms at night and under what conditions stoppage continues. Focus more on unresolved risks than the PoC period.
Conclusion
CNC tending for existing machines is not just about installing a robot—it’s about connecting the states of the machine, workpiece, and people. Before ordering, measure actual loading/unloading wait, clarify the seven control states and safety/recovery conditions, and estimate supply, gripping, chip handling, and measurement as a single cell. Test exceptions in FAT and SAT, and use good part count and stoppage reasons in a 90-day PoC to validate your investment hypothesis. This sequence reduces post-installation rework and additional site work.
If you are considering automating existing CNCs and want to start by checking door/chuck signals or organizing your RFP, contact TOMAS TECH with your current machine models and target parts. Even before finalizing specifications, you can consult on how to organize your survey items.
References
- Haas Automation, NGC Robot and Automation / Robot Integration Aid, Revision B 09/2026: External robot integration requirements and signal examples for Haas NGC machines. Reference as model-specific information.
- FANUC America, Machine Tending Robots: Use cases for loading/unloading, post-process transfer, measurement, etc.
- FANUC America, 10 Questions to Ask Before Automating Machine Tending: List of questions for consideration.
- ISO, ISO 10218-2:2025: Standard for safety requirements of industrial robot applications and cells.
- OSHA, Machine Guarding: General Requirements: US reference material on machine guarding. Not a substitute for Thai law.
- Universal Robots, Cobots in Machine Tending, 2023: Reference on design elements for machine tending.
