Robot Maintenance Support 2026 — Downtime Is Decided by Where the Parts Are, Not the Contract Tier

Which hours does robot maintenance support actually cut?

When you compare robot maintenance support offers, the number every proposal puts at the centre of the page is response time. Next business day, same day, or 24 hours a day and 365 days a year. Yet once you break down the hours of a day when a robot actually stopped, the share that the contract tier moves turns out to be startlingly small. The reason is simple. A maintenance contract promises how many hours until someone arrives, not how many hours until the line runs again.

Three things decide the time between a robot stopping and production restarting. Where the part physically is. Whether the information needed to restore the cell still exists. Who on site is allowed to touch the machine. At an overseas plant these three account for roughly 80% of MTTR (mean time to repair). The contract tier acts on part of the remaining 20%.

This article follows one model case all the way to the end in numbers — a Japanese-owned automotive parts plant in Rayong, Thailand, running 12 articulated robots — and asks how many hours a proper maintenance setup removes, what it costs, and how many years it takes to pay back. Four conclusions up front.

First, MTTR today is 52.0 hours per event, and 42.0 hours of that, 80.8%, is waiting for a part. What the maintenance contract shortens is mainly the 3.0 hours of fault isolation, which is only 5.8% of the 52.0 hours. Second, once maintenance is properly built, annual downtime falls from 468 hours to 232 hours, a reduction of 236 hours, and the loss falls by 1,298,000 baht. Third, of the 489,000 baht in upfront cost, 465,000 baht (95.1%) can be outsourced, and the 24,000 baht (4.9%) that cannot be outsourced is the cheapest and the most effective money in the whole plan. Fourth, and most importantly, a plant that has only one unplanned stoppage a year will never recover the 285,000 baht spent on local spare parts stock. The payback period comes out at 78.9 years. So the first thing to do is not to sign a contract or buy stock, but to count how many unplanned stoppages you had and how many hours each segment took.

For background, here is one figure worth stating and then setting aside. The robot density published by the International Federation of Robotics (IFR) is 132 units per 10,000 employees as a world average, 131 in Asia, 267 in Western Europe and 204 in North America. By country, Korea is at 1,220 units, Singapore at 818, Germany at 449 and Japan at 446. That release does not include a figure for Thailand, so this article does not discuss Thai density. What the number shows is only a direction of travel — robots are ceasing to be special equipment. Density itself answers nothing about how you should organise maintenance at your own site. That answer comes from your own fleet, your own configuration, and your own record of stoppages.

Split MTTR into five segments and the contract moves only 5.8%

Let me set out the model case. A Japanese-owned automotive parts maker in Rayong, Thailand. 12 articulated robots — 6 for arc welding, 4 for handling and 2 for palletising. Operation is two shifts, 16 hours a day, 250 days a year, which multiplies out to 4,000 hours a year. Average age since installation is 7 years, and 3 of the robots are past 8 years. The effective hourly cost of a maintenance engineer is taken as 320 baht.

Unplanned stoppages over the last 12 months at this plant looked like this.

The average rate is (4 × 9,000 + 5 × 2,700) ÷ 9 = 49,500 ÷ 9 = 5,500 baht per hour. It is an approximation weighted by event count, and in reality the rate moves up or down depending on which process stops. Even so, every calculation from here on uses that single 5,500 baht per hour figure. If you change the rate case by case, you can no longer read which measure actually worked. How much that simplification flatters the conclusion is tested later in this article, by putting the worst possible combination against it.

Segment-by-segment measurements

Measuring the span from stoppage to production restart across five segments — detection, fault isolation, parts arrival, recovery work, and accuracy verification — produces the table below. The “after” side splits into two columns, because the result is completely different depending on whether the failed part is one you can hold in local stock or not.

Reading that table vertically shows you what maintenance support proposals are not looking at. Of the 52.0 hours today, 42.0 hours is parts arrival. 42.0 ÷ 52.0 = 80.8%. Meanwhile, the segment that a maintenance contract clearly shortens is fault isolation, from 3.0 hours to 1.0 hour, a reduction of 2.0 hours. The measured improvement is 2.0 ÷ 52.0 = 3.8%. Even if isolation time could be driven to zero entirely, that is 3.0 ÷ 52.0 = 5.8%. In other words, moving the contract up one tier gives you at most 5.8% of the total to work with.

There is a harder fact behind that. Only 5 of the 9 events involved parts that local stock can cover — stoppages caused by batteries, cables, grease and sensors. The remaining 4 were caused by reduction gears, servo motors and control boards, all of which carry a high unit price and a high stock risk, so holding them locally cannot be justified. For those 4 events, parts arrival stays at 42.0 hours even after the maintenance build-out. The total is 48.0 hours, only 4.0 hours less than today’s 52.0 hours. If a proposal tells you that every event drops to 8.0 hours after the build-out, it is ignoring the parts you cannot stock.

On an annual basis, the picture is as follows.

Here is the arithmetic so you can follow it. Currently 9 × 52.0 = 468 hours, and 468 × 5,500 = 2,574,000 baht. After the build-out, the 5 stocked events give 5 × 8.0 = 40 hours and the 4 non-stocked events give 4 × 48.0 = 192 hours, totalling 232 hours. 232 × 5,500 = 1,276,000 baht. The difference is 468 − 232 = 236 hours and 2,574,000 − 1,276,000 = 1,298,000 baht. Averaged per event, 1,298,000 ÷ 9 = 144,222.2, and from here on it is rounded and handled as 144,200 baht. That rounded per-event figure carries straight through into the sensitivity analysis in the second half.

Now let me put a hole in my own numbers. The loss rate splits the events into “4 full-line stoppages and 5 cell-level stoppages”. The stock question splits them into “4 not stockable and 5 stockable”. These two four-and-five splits are different cuts and do not refer to the same four events. But in the worst case, they could overlap completely — the expensive 4 that shut the whole line could be exactly the reduction gears, servo motors and control boards you cannot stock. On the current side all 9 events run 52.0 hours, so splitting the rate changes nothing: 4 × 52.0 × 9,000 + 5 × 52.0 × 2,700 = 2,574,000 baht. Only the after side shifts. The loss after the build-out becomes 4 × 48.0 × 9,000 + 5 × 8.0 × 2,700 = 1,728,000 + 108,000 = 1,836,000 baht, and the saving falls to 2,574,000 − 1,836,000 = 738,000 baht. Using a single 5,500 baht per hour rate smooths that worst-case overlap flat. How the overlap changes the investment decision is picked up again in the sensitivity section.

Why parts arrival dominates at an overseas site

A parts arrival time of 42.0 hours looks absurdly long to anyone used to a plant in Japan. But this is not vendor negligence. It is structural.

First, where the stock sits. Major components such as reduction gears and servo motors are held at the manufacturer’s regional warehouse or at a plant in Japan. What the local distributor keeps is consumables and a limited range of units. If it is not inside Thailand, you are already talking about air freight.

Second, transport and customs. Air freight itself is fast, but the pickup cut-off time, the number of flights, customs clearance on arrival, bonded and permit checks, and finally the road leg to the plant all stack up. Stop on a Friday evening and that stack runs across the weekend. The 42.0 hour figure is an average that blends stoppages occurring midday on a weekday with stoppages occurring at a weekend. It is not a worst case.

Third, who decides. To order a part, you first have to establish that this specific part has failed. You cannot place the order until isolation is finished, so the 3.0 hours of fault isolation sit in series in front of the 42.0 hours of parts arrival. That is also why the one segment a maintenance contract genuinely improves has so little effect on the total. Cutting isolation by 2.0 hours changes nothing that anyone can feel if 42.0 hours are still waiting on the other side.

Which fixes the order of operations. Decide which parts to keep on site before you upgrade the contract tier. And to decide which parts to keep, you need to know what caused your past stoppages and how many there were of each. This way of thinking applies to maintenance work generally, and the wider equipment framework is set out in how to choose an equipment maintenance management system.

Industrial robot maintenance splits into three units of management

The biggest reason industrial robot maintenance plans drift out of alignment on the shop floor is that there are three different units in which due dates are managed, and people load all of them onto a single unit — the annual plan — without noticing.

Telling calendar years, elapsed years and servo-on hours apart

The test for telling them apart is what is driving the wear.

A battery discharges whether the robot is moving or standing still, so it is managed on calendar years. Grease is generally cited at around 3 years, counted from the date of the last change. The consumables list usually runs to batteries, brakes, reduction gears and encoders, and for every one of them, replacing after failure means the 42.0 hours of parts arrival get added on top.

The third one is the problem. Overhaul guidance is expressed in cumulative servo-on hours, and calendars cannot manage it. The model case plant runs two shifts, 16 hours a day, 250 days a year for 4,000 hours a year, but if another line in the same company runs one shift, that is 2,000 hours a year. Two robots of the same model installed in the same year will reach the threshold at points twice as far apart in time, because the one-shift machine takes twice as long to get there. They sit on the same row of the annual plan while one of them is actually wearing out at double the rate. That is why a maintenance plan built purely on a calendar will always drift.

Overhaul guidelines and the two-to-one gap between one shift and two

Mitsubishi Electric System & Service expresses its overhaul guidance in cumulative servo-on hours. Converting that back into years using our own assumption of 4,000 operating hours a year gives the following.

This is easily misread, so let me be explicit. The “6 years” and “9 years” are not figures published by the manufacturer. They are our own arithmetic, 24,000 ÷ 4,000 = 6 and 36,000 ÷ 4,000 = 9, using the 4,000 hours a year assumption of our model case. A one-shift machine (2,000 hours a year) would give 12 years for the body and 18 years for the controller, and on a three-shift operation the threshold would arrive far sooner. The moment you convert something the manufacturer states in hours into years, the number only means anything together with the assumption behind it.

Of the 12 robots in the model case, 3 have already passed the 24,000 hour guideline for the body. With an average age of 7 years at 4,000 hours a year, that is 28,000 hours, so it is hardly surprising. What matters here is whether anyone is reading out and recording cumulative servo-on hours at all. At a plant that cannot read them, nobody knows those 3 robots exist. They find out on the day one stops.

After the five-year statutory useful life, maintenance design starts to matter

The statutory useful life of an industrial robot is generally taken as 5 years. But that is a depreciation period under tax law, not the life of the machine. In practice there are many robots still running after more than 10 years, and fully depreciated robots remaining in service as core production equipment is the normal state of affairs.

That gap matters for how you design maintenance. For the first 5 years after installation, the manufacturer’s warranty overlaps with the early stable period, so unplanned stoppages are relatively rare. Maintenance discussions get deferred with “it is still new”. Then, around the time depreciation ends, the consumable replacement cycle comes round for the first time, the overhaul threshold approaches, and the engineers who commissioned the line transfer or leave. Weak maintenance surfaces after the book value hits zero.

Awkwardly, that is also the moment when an investment case is hardest to get approved. A request to put 489,000 baht into fully depreciated equipment is harder to explain than an investment in a new line. Which is precisely why it has to be written in downtime hours and loss figures rather than in impressions. In the model case, the question becomes how to position 489,000 baht upfront and 138,000 baht of additional annual cost against equipment that is currently generating 2,574,000 baht of loss a year.

For depreciated equipment, the question “should we replace it or strengthen maintenance?” always comes up as well, and the order is the same there. Only once you have the numbers — 468 hours and 2,574,000 baht a year — can you put a replacement case and a maintenance case on the same footing.

The most expensive part of robot breakdown response is lost information, not parts

Everything so far has been about the 9 unplanned stoppages. But the single most expensive event in robot breakdown response sits outside those 9. It is the event where the information needed to restore the cell no longer exists.

Say you replace a reduction gear. The part arrives, you fit it, you power up. Up to that point it is within the maintenance vendor’s scope of work. But after that, there is no guarantee the robot will trace the same path and lay the same quality of weld bead as before. The origin shifts, the tool coordinates change, the effective position of taught points moves. What saves you here is a backup of the teaching data. Without it, you re-teach from scratch.

And that work is outside the maintenance contract. A maintenance contract defines parts replacement, recovery work and inspection. It does not define “reproducing the welding conditions for your company’s products”. Even if you can get the vendor to do it, it comes as a separate quotation billed at an hourly rate.

What one re-teaching job at 372,800 baht is made of

If re-teaching a single arc welding robot takes 40 hours, the cost has two components.

The ratio is what you should be looking at. Labour is 12,800 baht, only 3.4% of the total. The remaining 360,000 baht is the loss from the line standing still. Re-teaching is expensive not because engineers are expensive, but because the line is down while it happens. Which is why the objection “we are fine, we can teach it ourselves” does not work against this number. If doing it yourself still stops the line for 40 hours, the 360,000 baht goes out just the same.

Meanwhile, the backup work that prevents this incident — Layer 2, described later — costs 9,600 baht. 372,800 ÷ 9,600 = 38.8. Prevent one single incident and it comes back at 38.8 times the outlay. No other item in a maintenance investment produces a ratio that clear. And yet this is the item that gets deferred most often, because a backup shows no visible benefit for as long as it is merely being taken.

The three backup targets and generation retention

There are three things to capture. Teaching data, parameters, and the PLC ladder logic around the cell.

• Teaching data. Taught points, tool coordinates, user coordinates, and the welding or handling condition settings. If even one robot is missing, that one robot has to be re-taught.
• Parameters. Origin data, axis settings, safety-related settings, communication settings. If these are missing, the taught points will not go to the same position even though they still exist.
• PLC ladder logic. Even if the robot alone is restored, the cell will not run if the handshaking with the surrounding equipment does not match. Plenty of plants take a robot backup only and assume they are covered.

The thing most often forgotten is generation retention. If you hold only the latest copy, you cannot go back when someone says “quality dropped after last week’s change”. Repeatedly overwriting the same file means almost the same thing as taking no backup at all. Date-stamp the files, keep several generations, and make it possible to see in one line which version is the current production condition. That alone changes the character of recovery after a parts replacement.

As for where to store it, a USB stick next to the robot is not enough on its own, because a fire or a theft takes both at once. That said, you do not need anything elaborate either. Create a folder per asset number on the plant file server, save with dates, and take a copy off site monthly. The 9,600 baht in Layer 2 of the model case is the cost of doing that job properly for all 12 robots.

Three types of maintenance contract, and where fleet size flips the answer

With that groundwork laid, let me organise the maintenance contract itself. Maintenance support offered by manufacturers and robot system integrators goes by different names at every company, but in substance it comes down to three types.

Spot, annual inspection only, and full maintenance

The three types differ in how the cost appears, how predictable the budget is, and how much work stays in-house.

What comparisons most often miss is that the same phrase “maintenance contract” describes completely different products depending on which of the three elements — scheduled inspection, parts and labour — it covers. Read the maintenance contract descriptions from Mitsubishi Electric System & Service or Omron and that separation is stated explicitly. When you compare quotations, before you compare the amounts, put on a single sheet whether parts are included, which parts if so, what the hourly labour rate is, and how out-of-hours and holiday work is treated. A comparison table that lines up only the prices is almost certainly comparing different things.

The cost picture for the 12-robot model case

The model case plant has used spot response until now. After the build-out, the assumption is an annual inspection contract covering all 12 robots.

216,000 − 78,000 = 138,000 baht. That is the increase in annual running cost. Per robot, 216,000 ÷ 12 = 18,000 baht a year for an annual inspection.

Why not go to full maintenance? The reason takes us back to the opening of this article. Moving up to full maintenance shortens mainly the fault isolation segment, and that is 5.8% of 52.0 hours. The 42.0 hours of parts arrival is decided by where the part is, not by the type of contract. There are plants for which paying a larger annual fee to make the budget predictable has real value, but that is an investment in flattening cost variance, not an investment in shortening MTTR. It is better not to compare two things with different purposes under the single phrase “maintenance support”. For a plant with a much larger fleet, or with a configuration where one robot stopping stops every line, the case for full maintenance improves. That threshold is set by fleet size and by whether work can be routed around a stoppage, not by the size of the price tag.

Separately from the type of contract, there is the question of who you ask. Manufacturer-affiliated, robot system integrator, or independent. How to make that choice is covered in how to choose a robot system integrator, and how to decide the scope of maintenance work to hand outside is set out in outsourcing equipment maintenance.

Robot preventive maintenance payback — stacking 489,000 baht across five layers

When you say you are going to “build out” robot preventive maintenance, the cost that actually arises falls into five layers. In-house work is converted at the effective hourly rate of 320 baht.

The five layers of upfront cost

All amounts in the table below are in baht.

The total is 14,400 + 9,600 + 285,000 + 84,000 + 96,000 = 489,000 baht. Cut that between in-house and outsourced and you get 14,400 + 9,600 = 24,000 baht in-house, 4.9% of the total, and 285,000 + 84,000 + 96,000 = 465,000 baht outsourced, 95.1%. By value it is overwhelmingly an outsourced project.

But what you should be looking at is not the size of the amounts, it is where the line falls. Layer 1 and Layer 2 cannot be outsourced even if you want to. How many servo-on hours your own robots have actually run, and which version of the teaching data is the current production condition, are things nobody outside your company can know. The clerical work of building a register can be delegated, but only you can judge whether the values going into it are correct. And that non-outsourceable 24,000 baht, 4.9% of the total, is the cheapest and the most effective. The arithmetic above — 9,600 baht in Layer 2 preventing one 372,800 baht incident — is the proof.

Layer 3, the 285,000 baht of local stock, only works for parts you can hold, as the breakdown makes clear. Reduction gears, servo motors and control boards are not in it. Layer 4 at 84,000 baht for special safety training and Layer 5 at 96,000 baht for procedures documentation are both investments in the question of who on site is allowed to touch the machine. The procedures are written in two languages, Japanese and Thai, because the people who write them and the people who read them are different people, and cutting that to one language means they never get used in practice.

Payback period and sensitivity analysis

The annual benefit is the 1,298,000 baht established earlier, and the annual cost increment is 138,000 baht. The net benefit is 1,298,000 − 138,000 = 1,160,000 baht, and dividing the upfront 489,000 baht by that gives 489,000 ÷ 1,160,000 = 0.42 years. About five months to payback.

Stop there and you have a beautiful proposal. But that 0.42 year figure hangs entirely on the assumption of 9 unplanned stoppages a year. So let us apportion by the per-event saving of 144,200 baht and vary the event count.

Each figure is 489,000 ÷ net benefit. At 6 events, 144,200 × 6 = 865,200, 865,200 − 138,000 = 727,200, and 489,000 ÷ 727,200 = 0.67 years. At 3 events, 144,200 × 3 = 432,600, 432,600 − 138,000 = 294,600, and 489,000 ÷ 294,600 = 1.66 years. Down to here it stands up perfectly well as an investment.

The problem is the last row. At a plant with one event a year, the saving is 144,200 baht. Subtract the 138,000 baht annual increment and the net benefit is 144,200 − 138,000 = 6,200 baht. 489,000 ÷ 6,200 = 78.9 years. In practical terms it never pays back.

And that 144,200 baht is itself an average across the split of 5 stockable and 4 non-stockable events. At a plant that stops only once a year, the result swings wildly depending on which side that one event lands. Even keeping the single 5,500 baht per hour rate, a stoppage caused by a stocked part saves (52.0 − 8.0) × 5,500 = 242,000 baht, while a stoppage caused by a non-stocked part saves only (52.0 − 48.0) × 5,500 = 22,000 baht. The latter falls below the 138,000 baht annual increment, leaving the plant 116,000 baht a year out of pocket.

Apply the worst-case overlap set out earlier — where the parts you cannot stock coincide with the expensive events that take down the whole line — and it gets harsher still. One event on a non-stocked part saves (52.0 − 48.0) × 9,000 = 36,000 baht, and even one event on a stocked part saves (52.0 − 8.0) × 2,700 = 118,800 baht. Neither reaches 138,000 baht. Under the worst overlap, the one-event-a-year plant is out of pocket either way. For completeness, applying the same overlap to the 9-event baseline gives a saving of 738,000 baht, a net benefit of 600,000 baht after subtracting 138,000 baht, and a payback of 489,000 ÷ 600,000 = 0.82 years. At a plant with real event volume, even the worst assumption does not change the direction of the conclusion. A payback period written from averages is least reliable at exactly the plants with the fewest events.

Do not delete the last row of the sensitivity table from the proposal. Giving a plant with one event a year 285,000 baht of local stock is an investment in letting parts sleep in a store room. What that plant should buy is the Layer 2 backup at 9,600 baht and the Layer 1 register at 14,400 baht, and nothing else. Together, 24,000 baht. Those two work regardless of the event count, because a backup pays off every time a part is replaced, even in a year with zero unplanned stoppages.

So the conclusion is not about stock or contracts. It is about sequence. Count the number of unplanned stoppages and the hours by segment before you make the investment decision. All the counting takes is the 45 hours of Layer 1, at 14,400 baht. There really are a fair number of plants deciding to buy 285,000 baht of stock without first paying that 14,400 baht.

Local issues specific to building robot maintenance in Thailand

The calculations so far work the same way at a plant in any country. A site in Thailand, however, has additional local realities around parts and around people. Neither shows up easily on a cost sheet, yet both move real MTTR substantially. I will also touch on the relationship with BOI measures, which comes up often when investment decisions are being made.

Import lead time and customs clearance for parts

In practical terms there are only three ways to shorten the 42.0 hour parts arrival time. Hold the part on site, confirm the distributor’s stock in advance, or make the procurement decision faster.

Holding the part on site is Layer 3, and some parts can be held while others cannot. The second, checking distributor stock, costs nothing. Make a list of the part numbers for reduction gears, servo motors and control boards, and find out once, before anything stops, how much of it the distributors or the manufacturer’s local subsidiary hold inside Thailand. Do that and the hours spent hunting for where the part is on the day of a stoppage disappear. The third is an internal matter. Even when the part is expensive, approval moves faster if everyone in the approval chain understands that the loss while the line is down runs at 5,500 baht an hour. It is a question of deciding, while nothing is down, who can approve up to what amount on the spot.

Deciding who is allowed to touch the robot on site

This is less about cost and more about organisation and regulation. Under Japan’s Ordinance on Industrial Safety and Health, teaching and related work for industrial robots is placed in Article 150-3, guarding during operation in Article 150-4, inspection and related work in Article 150-5, and pre-teaching checks in Article 151. The article on inspection requires that operation be stopped, that the start switch be locked out or fitted with a warning plate, and that a “work in progress” indication be displayed. As for special safety training, Article 36 item 31 of the Ordinance, made under Article 59(3) of the Industrial Safety and Health Act, covers teaching and related work, and item 32 covers inspection, repair and adjustment. In other words, not only teaching but inspection, repair and adjustment fall within the scope of special safety training.

Let me be straightforward here. Japan’s Industrial Safety and Health Act and its Ordinance do not apply directly to a plant in Thailand. Requirements on the Thai side are set out in the Factory Act and the relevant ministerial regulations of the Ministry of Labour, and those need to be checked separately. The reason the Ordinance is discussed here anyway is that the internal safety standards of Japanese-owned companies, and the audit checklists used by their head offices, are in practice very often drafted on top of it. Being asked at a head office audit who holds special safety training certificates is a situation that genuinely occurs. The realistic approach is therefore to check two things separately — the local statutory requirements, and the level demanded by head office standards. Do not draw a conclusion from only one of them.

On top of that, the organisational point. Deciding to outsource maintenance does not remove the need for people in-house who can touch the machine. Daily checks, witnessing vendor work, first-line fault isolation, restart and accuracy verification all stay inside. That is why the model case carries 84,000 baht in Layer 4 for special safety training for 4 staff. The figure of 4 is set as the minimum structure that secures at least one trained person on shift at all times across two shifts while still surviving holidays and resignations. One person per shift means the structure disappears on the day that shift’s engineer takes leave. Since engineers changing jobs is a realistic assumption at a Thai site, you keep multiple certified people, and you put the procedures into the 96,000 baht of Layer 5 documentation rather than into one individual’s head. The two go together.

BOI measures and equipment renewal

Applications to the Thailand Board of Investment (BOI) in the first half of 2026 were reported at 1,299 projects worth 1.473 trillion baht, up 37% year on year. Of those, measures relating to smart and sustainable industry accounted for 132 projects and 17.158 billion baht. It is a fact in itself that investment is moving towards automation and labour saving at a meaningful scale.

But reading “so maintenance must be incentivised too” out of those numbers is a leap. The scope and conditions of the measures differ from project to project, and maintenance spending on existing equipment is not necessarily eligible as such. What can be said is only this — if you are already considering equipment renewal or expansion, there is room to fold the maintenance build-out into that plan. As the number of newly installed robots grows, the Layer 1 register and Layer 2 backups can be built correctly from day one. Getting the new units right is cheaper than retrofitting the process onto the existing 12. The cost structure of new installations including collaborative robots is covered in the cost of introducing collaborative robots.

Pre-implementation checklist

Before you request quotations for maintenance support, check whether you can fill in the following items yourself. Comparing quotations while items are still blank means the things you are comparing are not aligned.

• How many unplanned stoppages occurred in the last 12 months. Did you count those that stopped the whole line separately from those confined to a single cell that could be routed around?
• For each event, did you write out the hours across the five segments — detection, fault isolation, parts arrival, recovery work and accuracy verification? If you cannot, is there a mechanism to record them starting from the next event?
• Can you state the loss per hour of downtime as a monetary figure, line by line? If not, who can calculate it?
• Can you read out cumulative servo-on hours for each of the 12 robots? Is there a register that records the values, and do you know how many robots have reached how far against the 24,000 hour body and 36,000 hour controller guidelines?
• Do backups of teaching data, parameters and PLC ladder logic exist for every robot? What is the date of the most recent capture, and are earlier generations retained? Is there a storage location other than the USB stick next to the robot?
• Of your past stoppages, how many could have been covered by local stock? Have you confirmed, before anything stops, the part numbers and unit prices of the parts that could not be covered, and whether distributors or the manufacturer’s local subsidiary in Thailand hold them?
• How many people on site perform daily checks and witness vendor work? How many hold special safety training certificates? Is that enough to survive holidays and resignations?
• Are the maintenance procedures and the recovery flow written in the language of the people who will actually read them?
• Does the quotation state clearly which of parts, labour, out-of-hours charges and call-out fees are included, and which are billed as incurred?

Frequently asked questions

How much does robot maintenance support cost per year?

It depends on fleet size and contract type. For the 12 robots in the model case, spot response averaged 78,000 baht a year in practice, switching to an annual inspection contract costs 216,000 baht a year, and the increment was 138,000 baht. Per robot that is a level of 18,000 baht a year for an annual inspection. But do not decide on that figure alone. Two contracts at the same annual price can be completely different things depending on whether parts are included and how out-of-hours work is handled. And this annual increment does nothing to the 42.0 hours of parts arrival.

Can we do industrial robot maintenance in-house?

It depends on the category. Daily checks, cleaning, teaching, grease changes, battery replacement and first-line fault isolation are handled in-house at many plants, provided the organisation and the training are there. Overhauls, replacement of reduction gears and servo motors, and control board replacement retain a dependence on outside support, because they need a procurement route for the parts and specialised adjustment. The important point is that deciding on an outsourcing scope does not make the in-house work disappear — daily checks, witnessing, record-keeping and backups all stay. And the Layer 1 register and Layer 2 backups that are the subject of this article cannot be outsourced in the first place.

When is a robot overhaul needed?

The guidance published by the manufacturer is expressed in cumulative servo-on hours — 24,000 hours for the body and 36,000 hours for the controller. When converting that into years, divide by your own operating hours. At the model case figure of 4,000 hours a year that gives 6 years for the body and 9 years for the controller, but those are our own calculations under our own assumption, not figures the manufacturer states in years. One shift (2,000 hours a year) doubles the number of years. Reading out your own cumulative servo-on hours comes first, and judging by the calendar comes last.

Which is better, manufacturer maintenance or a robot system integrator?

There is no single answer. Manufacturer-affiliated providers tend to be stronger on parts availability and model-specific knowledge, while system integrators tend to be stronger at restoring the cell as a whole, including peripheral equipment, PLCs and jigs. The dividing line is whether your past stoppages were failures of the robot itself or mismatches across the cell. If the former dominates, a manufacturer-affiliated provider fits. If the latter dominates, you want a partner who can see the whole cell. Which, again, you cannot determine unless you have been counting your past stoppages.

How long do robot spare parts take to arrive in Thailand?

It varies enormously by part type and by where the stock sits. In the model case measurements, parts not held in local stock took an average of 42.0 hours to arrive. Consumables such as batteries and cables that distributors carry can arrive the same day or within a few days, whereas parts such as reduction gears and control boards that have to come from a regional warehouse or from Japan run into several days once a weekend is involved. Local stock brings this down to 2.0 hours, but only for parts where the unit price and risk justify it. The recommended first step is to map out once, for your major components, how much stock actually exists inside Thailand.

Summary

When we compare robot maintenance support, we instinctively line up contract terms. But of the 52.0 hours of MTTR measured in the model case, what the contract tier can move is the 3.0 hours of fault isolation, 5.8% of the total. What governs the rest is the 42.0 hours of parts arrival, 80.8%. Downtime is decided not by the contract, but by where the parts are, whether the information exists, and who is on site.

Building maintenance out properly takes annual downtime from 468 hours to 232 hours, a reduction of 236 hours, and cuts the loss by 1,298,000 baht. Of the 489,000 baht upfront, 465,000 baht can be outsourced, but the 24,000 baht that cannot be outsourced — 14,400 baht for the Layer 1 register and 9,600 baht for the Layer 2 backups — is what actually works. The 9,600 baht of backups returns 38.8 times over if it prevents a single 372,800 baht re-teaching job.

And finally, the inconvenient conclusion from the sensitivity analysis, once more. At 9 events a year, payback is 0.42 years. At 6 events, 0.67 years. At 3 events, 1.66 years. But at a plant with one event a year, only 6,200 baht of net benefit remains, and payback of 78.9 years means it never happens. Stock is not a universal answer. So the sequence starts not with a contract and not with stock, but with counting the number of unplanned stoppages and the hours by segment. Layer 1, which does the counting, is 14,400 baht and 45 hours of work. An investment decision that skips it is a bet on whether you happen to be right.

It is fine if you do not have the numbers yet. Not knowing how many stoppages you had, not knowing how to read out cumulative servo-on hours, not knowing how far your backups actually go — starting from that state is the situation we are asked about most often. As a step before quotations, we can help you map out your current segment times and event counts together, so please feel free to get in touch through our contact page even while you are still at the review stage.

References

• IFR press release on robot density (International Federation of Robotics. World average 132 units per 10,000 employees, Korea 1,220, Singapore 818, Japan 446. No figure for Thailand is included in this release)
• Thailand BOI investment applications in the first half of 2026 (Nation Thailand, 23 July 2026. 1,299 applications worth 1.473 trillion baht, of which smart and sustainable industry measures account for 132 projects and 17.158 billion baht)
• Commentary on safety regulations for industrial robots (Japan Quality Assurance Organization. Teaching and related work in Article 150-3, inspection and related work in Article 150-5, special safety training in Article 36 items 31 and 32 of the Ordinance) and an overview of how the articles are arranged (Rodo Shimbunsha)
• Statutory useful life versus actual years in service (Kyoni. The 5-year statutory useful life is a depreciation period under tax law, and many robots are reported to run for more than 10 years)
• Robot overhaul (Mitsubishi Electric System & Service. Guidance is 24,000 hours for the body and 36,000 hours for the controller. The conversion into years is our own calculation)
• Robot maintenance contracts (Mitsubishi Electric System & Service) and maintenance contract services (Omron. Both make clear the separation between scheduled inspection, parts and labour)
• Robot consumables and replacement guidelines (MIRAI-LAB. Grease changes at around 3 years, consumables being batteries, brakes, reduction gears and encoders)