In January 2023, I was standing next to a small packaging machine that had gone quiet. The smell of burnt plastic hung in the air. I opened the gearbox and saw what was left of an mg995 servo motor that I had selected for a retrofit project. The meltdown wasn't a surprise—I'd been hearing the high-pitched whine for two days. But it was still a gut punch.
I've been designing motion systems for industrial equipment for about 12 years. In that time, I've made my share of mistakes, and I keep a written record so I don't repeat them. This incident was mistake #12 on that list. The total cost of that retelling, including the replacement hardware and the time I wasted, was roughly $8,500. Looking back, I should have known better. I had the right brand in front of me all along—thomson-linear had the product, and I ignored it because it wasn't the cheapest option.
How I Tried to Cheapen the Bid
It started with a customer request. They ran a packaging line for a food company, and one station used a pneumatic cylinder with a fixed stroke. They wanted to make it programmable: multiple stroke lengths, adjustable speed, and better position repeatability. They asked me to quote an electric alternative.
My first instinct was to quote a complete thomson linear actuator with an integral motor and screw. I'd done dozens of those projects, and they always worked. But the price came back around $3,200, and the customer—accurately—said that was over half of their equipment budget. So I decided to be clever. I told them I could build a similar system from parts for about $1,400.
Here's what I put on the BOM:
- One surplus thomson linear shaft (20 mm diameter, 600 mm long)
- Two open linear ball bearings with rolled ball cages
- A 16 mm ball screw from a surplus dealer, no documentation
- An mg995 servo motor that I had in the shop (left over from a robotics demo)
- A cheap stepper/servo driver with a current limit knob I didn't fully trust
I remember thinking: "The MG995 claims 9.4 kg·cm of torque at 4.8 V. The moving mass is only about 3 kg. This will fly." (That sentence should have been a red flag. I was designing around a hobby motor, not an industrial one.)
What I Missed (and What a Ball Bearing Taught Me)
The failure wasn't immediate. The system moved the way I expected on the bench. Then the customer installed it on the line, and within two hours the motor was getting hot. I checked the current—it was spiking to nearly three times the MG995's rated continuous value. I also started hearing a rattle from the ball screw nut, something that hadn't shown up at low speed.
That noise led me to a deeper lesson about how ball bearings are made. I used to think a ball screw is just a threaded rod with balls stuck inside—no big deal. That's a legacy myth from an older manufacturing era, when most machinery used sliding friction and a precision ball screw was a rare luxury. Today, a re-circulating ball screw is essentially a precision ball bearing wrapped around a threaded shaft. How ball bearing made matters enormously: high-quality balls are ground and polished from bearing-grade steel, the raceway in the nut is honed to a controlled arc, and the internal clearance is set to microns. A cheap surplus screw, by contrast, may have rolled threads with rough surfaces. At low speed, it feels fine. At higher speed and load, the difference shows up as heat, noise, and wear.
But the biggest mistake wasn't the screw. It was the motor.
The MG995 is not an industrial actuator. It's a hobby servo, designed for intermittent rotation in RC cars and robot arms. Its datasheet shows a stall torque of 9.4 kg·cm, but that's an instantaneous value, not a continuous rating. Industrial motors are specified with a thermal limit and a service factor; a hobby servo has none. I didn't check the duty cycle, and I didn't calculate the motor-to-screw inertia ratio. The controller was trying to correct every micro-vibration, which made the motor run hot even when the load was low.
The most frustrating part is that I had all the information in front of me. The thomson-linear website has duty cycle curves for every actuator. The NEMA MG1 standard specifies thermal performance for AC motors. I could have downloaded a datasheet for a proper servo motor. Instead, I trusted a $9 hobby motor that was never meant for eight-hour production shifts. (Ugh.)
The Burnout
At 3:17 PM on the second day, the MG995 finally stopped responding. The customer called me, and I drove over to find the plastic gearbox literally melted. The line was down for the rest of the day. The customer didn't say much. They didn't need to. I'd recommended the cheap solution, and it had cost them real money.
I spent the weekend redoing the quote. This time, I did the math properly. I selected a real thomson linear actuator with a brushless DC motor, rated for continuous duty at 40% cycling. It cost more, but it came with a mounted encoder and an engineered screw assembly. I paired it with a century ac motor and an inline gearbox that we had in the warehouse for an auxiliary fan drive. The Century motor was the right size for the load, had a service factor of 1.15, and was already wired for the plant's 480 V supply. The actuator did the positioning, and the AC motor provided the vertical lift through a toothed belt. It worked—not because the hardware was fancy, but because each component was within its specification envelope.
In the post-mortem, I realized the true cost of my cheap BOM: $1,400 spent on the first attempt, another $2,300 for the replacement parts, $1,800 in labor and downtime, and about $3,000 of my engineering time. The customer paid for the machine twice, and I got a reputation hit that I probably deserved.
Doing It Right, Finally
My current checklist starts with three questions:
- What is the actual duty cycle and speed profile?
- What is the load-to-motor inertia ratio?
- Can the drive train handle the required accuracy and repeatability?
Only after I answer those do I look at brands and prices. If the application calls for continuous operation, a hobby servo is out. If the customer needs a package that assembles in under three days, a modular thomson linear actuator is often a better choice than a custom stack of parts. If the plant already has single-phase power, a Century AC motor can be a solid choice; if they run three-phase, a motor sized to the load is cheap enough.
I've also learned to say "this isn't right for you" more often. There is no universal best linear actuator. An MG995 is fine for a robot arm that moves occasionally—just not for a production line. A thomson linear shaft is great for a simple guided rod, but if you need zero-stick-slip motion at high speed, use a linear bearing system. A ball screw with accuracy class C7 is fine for a feed table; for a CNC spindle, you need class C3 or better. Honesty about these boundaries is what builds trust, and it's a lot cheaper than writing a refund check.
The Checklist I Now Use Before Every Linear Motion Spec
I keep this list on the wall above my desk, and I run every new project through it. Could it prevent every mistake? No. But it has caught 47 potential errors in the past 18 months, which saves real time and money.
- Duty cycle: Compare the motor's continuous torque against the RMS torque of the actual cycle.
- Inertia ratio: Keep the reflected load inertia within 10:1 of the motor rotor inertia for good control response.
- Ball screw quality: Look for at least ISO 3408-3 class C7 for general automation; use C5 or better for precision positioning.
- Shaft support: Align the linear shaft or leadscrew within the manufacturer's stated tolerance; misalignment kills bearings.
- Thermal limits: Check that the motor and screw can dissipate the heat from the duty cycle; consider ambient temperature.
- Lead time: If the customer needs it in a week, don't design a custom stack of parts—buy a tested actuator.
I'm not saying you should never build from components. For a one-off R&D machine, a combination of a thomson linear shaft, a quality ball screw, and an industrial motor can be extremely cost-effective. But when the machine is going to run for years, the long-term cost of a failed drive is always higher than the short-term premium for a matched system.
The smell of burnt plastic finally taught me that lesson. I just wish it hadn't taken $8,500 for me to listen.