Digital X-Ray Detector PCB: Readout Noise, Power Integrity and Production Consistency for Flat Panel Detectors
Flat panel X-ray projects rarely fail at the scintillator or TFT but
Data Center Cooling Fan Controller PCB: Heavy Copper, Hot-Swap Redundancy and TACH Signal Integrity for Fan Walls
- sprintpcbgroup
A Fan Wall That Kept Losing Speed
Not long ago I squatted in a machine room for nearly a week, just to investigate a fan wall that kept dropping speed for no apparent reason. At first I thought the control algorithm had not been written precisely enough, and after changing several versions of PID parameters I found it was not that at all. The problem was in the board itself: the current loop on that Data Center Cooling Fan Controller PCB had an outrageous voltage drop at high temperature, and the 12 volt bus fell to just over ten volts by the time it reached the end. The fans were still turning, but the measured speed feedback was already shaking badly, and the BMC side directly reported a pile of false faults. This made me reconsider a fairly blunt principle: many people discussing cooling systems like to stare at speed control strategy and smart algorithms, but what truly chokes the throat is often the neglected copper thickness and stackup structure.
Superstition About High-End Schemes
We used to be superstitious about so-called high-end schemes too, forcing the drive topology, sensorless sine wave and FOC all onto one MCU, and the boards that came out had a mess of signal integrity. Later I directly separated out the driver stage and honestly found a heavy copper PCB manufacturer that could make both thick copper and multilayer mixed lamination, placing the whole power loop on the 3 ounce copper inner layers, with the top and bottom layers carrying only signals and auxiliary power. After this adjustment, the half-load temperature rise that used to occur dropped by a dozen or so degrees. This also made me more careful in choosing a multilayer pcb supplier: not everyone can press a ten-layer board with thick inner copper without delamination or blistering, and many manufacturers say with their mouths that they can do it but expose themselves after one pass through reflow soldering.
HDI and Heavy Copper Together
On the multilayer side, there is another point that is rather counterintuitive. Now there are more and more calls for HDI in making data center fan control boards, because board size is pressed dead by structure while circuit complexity keeps rising. I have contacted several hdi pcb manufacturers, and some can indeed make buried and blind vias very beautifully, but once powered up they cause trouble: via resistance on the high-current path is uneven, causing the current distribution of multiphase fans to skew badly. Later I simply switched to an hdi pcb supplier that could do both HDI stackup and heavy copper, having them fit laser drilling and thick copper into one process window. The price went up, but at least I would not be called awake by operations in the middle of the night.
Putting Thermal Logic and Signals on the Board
On speed regulation, I now prefer to push the temperature control logic down into onboard hardware rather than counting on the chassis management controller to compute it. Many people feel high-speed signals matter only on communication links, but in fact for the PWM frequency modulation and multichannel TACH sampling of a fan control board, once you get to a high-power multi-fan array, noise coupling is a real headache. So I later specifically found a high speed PCB manufacturer familiar with high slew rate signal routing, who ground out everything from interlayer isolation to return paths bit by bit, and only then pressed the edge jitter of the speed signals down to an acceptable degree.
In the end, a fan wall in a data center is a typical weakest-link system. However good a fan you can buy, if the control board cannot even keep its basic heat dissipation copper, then the whole design of speed regulation, redundancy and hot swap turns into a decoration. When I chat with the team about a new scheme now, my opening line is always one question: is there enough copper on the board, and dare you open the stackup for me to look at myself? That may be more practical than discussing any model predictive control.
Where the Heat Really Comes From
Honestly, after more than ten years making data center fan control boards, my biggest takeaway is that many people think this matter too simply. They think that finding a reliable HDI PCB manufacturer to make the board, choosing a driver IC and tuning the PWM duty cycle in the program makes the fan run, but it is not like that at all.
The problem lies in heat generation. Not chip heat, but copper foil. You stuff dozens of servers into a cabinet eighteen units high, with six to eight brute-force fans in each server, and running at full speed the current can reach three or four amps at twelve volts, which works out to forty-some watts for a single channel. If the board uses conventional one ounce copper, the traces carrying power quickly become small heating wires of several hundred degrees and then burn through. We paid for this early on: the multilayer PCB supplier we used made it to consumer standards, and it did not survive two weeks of stress aging testing. Later we turned to a factory able to make Heavy copper PCB and pulled copper thickness directly to four or even six ounces, with inner layers carrying the high-current bus and outer layers carrying signals, and only then was temperature rise held down. But heavy copper brings a new problem: the etching precision of thick copper is not as good as thin copper, line width and spacing must be left very large, and with the same area fewer signal layers can be routed. So we were forced to go HDI, using laser microvias to fan lines out to the thin copper layers and then using buried and blind vias to interconnect with the thick copper layer. At this point what is tested is the multilayer PCB manufacturer’s lamination alignment capability: the expansion and contraction coefficients of a thick copper core and the HDI build-up differ, and after pressing the board easily warps and causes cold solder joints in SMT. One batch of ours had lamination parameters set wrong, causing the BGA solder balls and pads to open slightly, so that fan speed feedback came and went and the monitoring backend alarmed like mad.
The Hardware Basis of Speed Control
This leads to the hardware foundation of speed regulation. Many people think PWM speed control is just a microcontroller sending a square wave and the drive MOSFETs switching on and off, but in a data center environment, switching noise runs along the ground plane into the neighboring I2C or UART bus, causing the system to misjudge the fan state. You must make the ground of the drive loop an independent star grounding point, with power ground and signal ground connected only at a single point at the DC/DC module at the board entrance. If this topology is not handled well in PCB layout, then even if you find the most capable high speed PCB manufacturer to do signal integrity simulation, it is useless. A high-speed PCB factory understands differential length matching and impedance control, but ground bounce under large current and strong interference needs solid power integrity experience, and this is not something any multilayer PCB supplier can guarantee.
Mixed Analog and Digital Reality
I often tell customers that the essence of a fan control board is a harsh application mixing analog and digital. The digital end runs only a few hundred kHz of I2C, which cannot be called high speed at all and does not require chasing board materials above H3, but the analog end is deadly sensitive to power ripple and thermal gradients. Take TACH speed feedback: that line is an open-drain output, so what pull-up resistor do you choose? Too small wastes power, and too large makes the waveform’s rising edge too slow and easily filtered out as a glitch by interference. In our second design, we placed the pull-up resistor within three millimeters of the controller and then had the HDI PCB supplier prototype a four-layer first-order sample board, sandwiching the TACH trace between two complete ground planes, and only then did crosstalk fall to an acceptable level. You will find that these details simply cannot be seen at the chip selection stage and were completely bought with the losses of more than ten layout revisions.
Redundancy and Hot Swap on the Board
The driver IC is instead the link least worth worrying about. The fan drivers selectable on the market now are very mature, with stall protection, over-temperature frequency reduction and phase current limiting all built in, and basically building by the reference design makes the function run. The real devil is in the redundancy architecture. In a scheme we made for a certain financial data center, a single fan control board was required to carry four redundant channels, with a short circuit on any channel unable to affect the other three, and with support for online hot replacement. To achieve this, each channel’s drive output had to be in series with an independent ideal diode controller, isolated with back-to-back N-channel MOSFETs, so that once the front stage shorts, the controller can cut off the gate within two microseconds. That is not all: if the inrush current at hot plug is not limited, it makes the twelve volt bus of the cabinet-level power supply drop instantly below nine volts, and the switch chip in the same power domain restarts directly. So we added a hot-swap controller with dV/dt control at the board input and used a heavy copper power bus bar for current collection, to avoid arcing at insertion and removal caused by excessive PCB trace inductance. These board-level measures cannot be settled by simply finding a Heavy copper PCB manufacturer to add copper thickness, and you have to plan the current path from the system perspective and then reversely constrain the PCB’s stackup and device layout.
Finally, a word about how hard intelligent speed regulation algorithms eat into hardware resources.
Sensorless FOC and the Fan Noise Problem
Some time ago a friend in data center operations came to chat with me about cooling fans. Their machine room has a batch of old servers still using square-wave-driven three-phase fans, which are fine at low load, but once speed is pulled up, that kind of high-frequency whine can be heard through the cabinet door. He asked whether swapping in a controller supporting sine wave drive would solve it, and I said you are in the right direction, but it is not that simple.
Using sensorless FOC on a server fan controller, the trouble has never been in the chip itself but in how to run the signals steadily on the PCB. More than ten years ago when I had just entered the field, I made several boards based on TI schemes, and at that time I felt that placing the op-amps well and positioning the sense resistors correctly settled everything, but when the prototype came back and was powered up, the phase current waveform had a frightening number of glitches, and although the motor turned, torque ripple was not much better than with square-wave drive. After repeated revisions I found the problem was at the board factory. The ordinary multilayer board supplier we used had uneven inner-layer copper thickness, and the interference of the power layer on the high-speed current feedback signal simply could not be pressed down. Later we changed to a factory accustomed to heavy copper PCB, pulling the power ground loop out separately with 2 ounce copper and combining an HDI structure to bury the small analog signals in inner layers, and only then did the waveform become clean.
Heat in a 2U Chassis
Finding an HDI PCB manufacturer able to make this kind of board is not hard now, but what is hard is that they can also take care of heat dissipation. With twelve fans running at full speed in a standard 2U chassis, even if the MOSFETs on the controller have on-resistance as low as milliohms, the accumulated heat is enough to push the PCB surface temperature past 90 degrees. The toughest design I have seen directly embedded a ceramic substrate into a multilayer board, which a volume-production multilayer PCB supplier simply cannot make, and you must find a factory that specializes in data communication orders, since only they have a process that can make copper thickness above 3 ounces without delamination.
Hardware Over Algorithms
Many people think that in making an FOC controller, the algorithm is the core and the hardware is just a carrier. I instead think that in a scenario like server fans, which demands high acoustic noise performance and reliability, whether you can find the right high speed PCB manufacturer is the watershed. If you ask an ordinary factory to handle EMI at a 25kHz switching frequency, at most it draws an isolation slot and adds a shield can, but a knowledgeable HDI PCB supplier will, from the start, suggest adjusting the stackup order to sandwich key signals between two ground planes, and calculate enough copper width for the high-current path, so that you do not have to worry about heat every day afterward. This depends not only on theory but on experience accumulated by honestly tumbling.
In the end, whatever sensorless, sine wave and FOC, what lands in the product is a palm-sized circuit board. Those with beautifully made parameters all have behind them a multilayer PCB manufacturer who can understand your needs, not the kind of factory that only builds according to the Gerber files and throws up its hands when problems occur.
Supply Chain Choices Behind PWM Accuracy
Over the last few years I have handled several data center cooling retrofit projects, and the biggest pit I stepped in was not the fan selection itself but the supply of those few PCBs of the fan controller. Many people think that once the PWM drive scheme is settled, the rest is finding any multilayer PCB manufacturer to prototype, but it is not like that at all.
Take a batch of high-power-density cabinet fan controllers we made last year: the board ran several PWM speed signals simultaneously plus current sampling and temperature feedback, and the interference was simply too large. At first, to save trouble, I found an ordinary multilayer board factory, and the result was that ground bounce noise directly ruined the PWM duty cycle accuracy, fan speed jumped around when load changed, and ILO alarms rang continuously. Later I honestly switched to a factory specializing as a high speed PCB manufacturer, readjusted the stackup and routing, and handled power ground and signal ground in separate zones of the multilayer board, and only then was the problem pressed down. After that I became firmly convinced of one principle: the fan controller PCB used in a data center has outrageously high requirements for thermal management and signal integrity and cannot be taken by just any multilayer pcb supplier.
Copper Thickness and Dimensional Pressure
Another underestimated detail is copper thickness. The instantaneous current of some high-power fan drive stages can rush to several dozen amps, and if the board is still drawn with conventional 1oz copper, temperature rise and voltage drop simply cannot hold. One revision of our control board used 4oz inner-layer copper from a heavy copper pcb supplier to keep trace width within a reasonable range, otherwise the board would have to be a size larger and could not fit into a 1U chassis at all. However, the process dispersion of a heavy copper pcb manufacturer is larger than that of conventional factories, and if the etch factor is not uniform, line width tolerance eats your design margin, so when choosing a supplier you must stare at their process capability data.
Size is another factor pushing process upgrades. Inside cabinets, space is getting more compact, and the hub board of a fan array often has to cover drive, monitoring and communication, with frightening component density. Traditional through-hole multilayer boards simply cannot spread lines out, and you have to rely on an hdi pcb manufacturer to raise routing density with microblind and stacked vias. I once used a 2-order HDI control board in a four-fan 2U module, with a board thickness of only 1.2mm but six layers of interconnect completed, fanning out 12 MOSFETs and a pile of resistors and capacitors, and without this process it simply could not have been made. Of course, there are not that many hdi pcb suppliers who can do this kind of board well, especially those who can stably control interlayer registration accuracy, and the trial-and-error cost is not low.
PWM Interface Conditioning
On PWM drive itself, many people like to add no treatment at the interface and directly slam the speed control signal coming from the BMC onto the driver IC pin. This practice may be fine in a quiet laboratory environment, but in an actual rack, common-mode noise coupled in by long cables can make the PWM signal unrecognizable. My habit now is to add a stage of hysteresis comparator and RC filter at the interface end of the board, shaping the signal into a clean square wave before feeding it into the drive stage, at the cost of only a tiny bit more PCB area but saving countless troubles of on-site debugging. The fan’s acoustic performance and bearing life are also much steadier, with no sudden speed changes caused by occasional glitches.
In the end, a board like the Data Center Cooling Fan Controller PCB looks simple but in fact is pushed to the limit of board-level process in several dimensions: signal integrity, current-carrying capacity and integration. Choosing the wrong factory can collapse the reliability of the whole cooling system, so later at the start of every project I would approach several multilayer pcb manufacturers and heavy copper pcb suppliers at the same time and evaluate their actual process capability from design drafts, rather than looking only at the numbers on the quotation. The little board fabrication money saved is not worth mentioning compared with online downtime.
A Small Board with Everything Inside
In these years handling several data center cooling projects, I have increasingly felt that the palm-sized fan control board has the character of a sparrow small but with all its organs. Many people think it is just driving a few fans, drawing a circuit and pulling a few wires and done. Only after stepping in pits do you know that for a reliable Data Center Cooling Fan Controller PCB, from material selection to stackup, from routing to finding the right supplier, every link hides a possibility of the project flipping over.
In my last project the fan power was not large, just a small turbo fan of about 12V/10W, but the board had to be stuffed into the side of a very flat 1U chassis, with size pinned dead. At that time a conventional FR-4 double-layer board simply could not route: power lines too thick would not fit, too thin would get hot, and signal lines also had to wind around. In the end I found a factory working as an hdi pcb manufacturer and used a first-order HDI process, laser drilling to bury the power layer into an inner layer, leaving only logic signals and connectors on the surface. Many people think HDI boards are expensive, but if you do the arithmetic, you understand: the area saved turns into larger airflow cross-section in the whole system, which is instead a plus for heat dissipation and is more cost-effective than adding more fans. And HDI lets the PWM speed control signal and TACH feedback traces be buried in inner layers, with the surface layer wrapped by a complete ground copper, so external electromagnetic interference basically cannot stick on, and the speed signal is as clean as if washed.
Why I No Longer Design Below Four Layers
Speaking of multilayer boards, I now almost never make a fan control board with fewer than four layers. Even a simple 4-wire PWM fan hub, I use a multilayer pcb manufacturer’s scheme. The two extra layers of ground plane and power plane are not only for routing convenience but because the fans in a data center run all year without rest, and problems such as floor noise, ringing and occasional EMI may not be obvious in early testing, but after running three months or a year they may cause some MOSFETs to fail inexplicably. The naturally low-impedance loop of a multilayer board lets current find the shortest path home, and even if the switching node layout is slightly reckless, it will not turn into a mess. When I meet an upper-bridge-driven high-current fan, such as an axial fan above 48V/30W, I even make the power layer 2 ounce thick copper and have a heavy copper pcb supplier laminate it specially. Inner layers carry the large current and outer layers carry logic control, with thermal vias densely drilled, so that the path of conducting heat to the metal housing is short enough. Some say the board becomes thick and heavy this way, but I think it is worth it, which is better than replacing fans on site and being penalized for downtime.
Supplier Experience From High-Speed Edges
Choosing suppliers is also experience I was forced to grow in these projects. Not every multilayer pcb supplier can digest the subtle points of a fan board. Once, to save trouble, I used a factory that did not make high-speed boards, and once the PWM frequency went up to 25kHz, the pulse edges were like sawteeth and the remote fans simply could not run steadily, with speed going high and low. Later I switched to a supplier that was also a high speed pcb manufacturer, who took impedance control much more seriously, and length-matched and differential routing of traces from the controller to the fan socket were no problem, and the waveform became clean at once. High-speed PCB process is not only for communication boards: now many fan driver chips use high-frequency sine drive, with carriers of tens of kilohertz or even higher and edges as steep as a knife cut, and without the high-speed board design approach, crosstalk between phases can kill your debugging.
Redundancy as Physical Isolation
Speaking of redundancy, that is an iron law of data centers, but landing on the PCB it is not only a sentence of hang one more fan. When making a board for a certain liquid-cooling auxiliary cooling system, the project required fans to support N+1 hot swap. The three words hot swap require board-level design to regard each fan channel as an independent power supply unit. My approach at the time was to use an independent load switch and fuse for each fan channel, with the power entrance pulled by wide copper separately to the 48V bus, never sharing a thin bottleneck section. In this way, even if one fan shorts internally, the short-circuit current at most blows its own fuse, the bus voltage does not move at all, and other fans keep turning as they should. This design looks to have no technical content at a glance, merely costing copper and area, but all the decency of automatic fault switching comes from the hard isolation of the power path. Without it, so-called redundancy is electrically a house of cards that falls over at a push.
On routing I have another insistence: the TACH speed feedback line in my eyes is like a sensitive little beast, and it must never be allowed near any PWM switching node. However tight the board space, I will pull the TACH line separately to a position far from the drive loop, sometimes even specially wrapping it in ground and adding shielding vias. Because the BMC chip on a server motherboard is extremely sensitive to fan fault judgment, a tiny glitch coupled onto the feedback line may be misread as a fan stall, and then the whole cooling strategy goes haywire and the machine room roars.
The Night a Hot-Swapped Fan Disturbed Storage
A few days ago, when chatting with old buddies from operations about the mess in the machine room, I immediately remembered that night when, because of hot swapping a fan, the storage of a whole row of servers acted up. The fan controller board we used at the time was a reference design copied by a small factory, a two-layer board with traces squeezed together like noodles. The problem was exactly at the instant of hot swap: the board’s ground bounce disturbed all the speed signals of adjacent fans, the management platform instantly flashed seven or eight alarms, and the young man on duty nearly pulled out the newly replaced fan again.
Later, calming down and opening up that Data Center Cooling Fan Controller PCB, I found that the ground return of power and signal had never been seriously separated. The current return loops of a dozen-plus fans were mixed into a pot of porridge on the board, and the surge pulse of dozens of amps produced by plugging and unplugging one fan directly sliced a string of voltage spikes on the common ground impedance. Glitches like this coupled onto the TACH pin made the speed the controller read look like an arrhythmia, with no way to judge whether the fan was broken or the insertion was bouncing. This matter made me thoroughly understand that for a hot-swap environment, designing a fan control board absolutely cannot use consumer-grade PCB thinking.
The Second Revision: Six Layers and Separate Returns
Our second revision scheme decisively changed the route. We found a factory specializing as a high speed PCB manufacturer to cooperate, and although the fan control signal itself cannot be called high speed, ground plane integrity and stackup structure are the basic skills of high-speed boards. It was changed to a six-layer board, completely separating the high-current power layer and the signal layer physically, with each fan occupying its own section of trace starting from the bus branch point and the return path no longer shared. At the same time, because the surge current peak was too fierce, we found another Heavy copper PCB manufacturer and increased the copper thickness of the power layer to 4 ounces, ensuring that the instantaneous current of capacitor charging at fan replacement would not burn hidden hazards into the copper. When this board was finalized, we internally required the supplier to be a factory with both multilayer pcb manufacturer and Heavy copper PCB supplier capability, otherwise it simply could not take our set of needs.
Interface Area HDI
Traces near the controller chip are the densest and the fan interfaces have to be concentrated, with space compressed severely, so for that small interface area we used the HDI process. We ran in with an hdi pcb supplier for three or four rounds before settling the stacked-via scheme of buried and blind vias and laser drilling, which both guaranteed the impedance continuity of the TACH and PWM differential pairs and did not pull the microvias apart through thermal stress. It is quite interesting to say: that hdi pcb manufacturer originally made mainly phone motherboards, and when they first took our data center order they felt the fan board had no technical content, but after the prototype was measured for signal integrity, even they were surprised that a hot swap of a fan of a few dozen watts could produce so many tricks.
Looking back now, that Data Center Cooling Fan Controller PCB is in essence a small power management platform mixing power and signal, which has to withstand surge, keep signals clean and also be plugged and unplugged ten thousand times without dropping packets. When choosing a multilayer pcb supplier you cannot look only at layer count and line width and spacing, and you have to see whether they have truly done a combination process with heavy copper mixed lamination and HDI. Many factories say they can do it, but the prototype comes back either with uneven copper thickness or with bubbles in via fill glue, and a board like this stuffed into a 7×24 non-stop cooling system is burying a mine for yourself.
So when peers ask me how to make a data center fan control board, I never say to refer to any reference design. I tell them directly: first find a PCB supplier that can do both heavy copper and multilayer high-speed mixed lamination, then use HDI to compress the interface area cleanly, and then do not be stingy about making power traces thick and independent where they should be, and this way hot swap is no longer the nightmare of the hardware team. After all, a fan looks simple, but it is the one component in the cabinet that is never allowed to stop turning, and if you make the PCB of its control board solid, the later system reliability can improve by at least two grades.
Copper Over Layer Count: A Cluster Lesson
I have tinkered with server cooling several times, and the pits stepped in while upgrading the cooling system of a computing cluster the year before last still make my feet ache when I think of them. Many people start by feeling that a data center fan control board must use so many layers of HDI board, almost wanting to pile a dozen-plus layers even on a palm-sized small board, and I feel this is a bit off track. It is not that multilayer boards are bad, and finding a reliable multilayer pcb manufacturer is of course important, but once in the machine room, whether the board can withstand long-term high current, whether it is easy to repair and whether the signal is stable are far more substantial than blindly chasing layer counts.
At that time we had a batch of self-developed fan control backplanes, and at first, to save space, used an 8-layer HDI design, with signal integrity indeed beautiful and TACH waveforms clean. For the first half year we used it happily, and later, when summer load came up and fan speed soared above 12000 rpm, the problem came. Copper thickness on the board was not enough, the power pins of the fan interfaces suffered long-term high current impact, and together with the slight vibration of backplane insertion, hidden fractures actually appeared. TACH feedback came and went, management software reported fan lost one moment and recovered automatically the next, and the brothers on duty in the machine room were tortured badly. This forced me to rethink: in an environment like a data center, however delicate a board you get an hdi pcb supplier to make, if current-carrying capacity and mechanical strength cannot hold, it is all decoration.
Splitting the Board: Heavy Copper Backplane Plus Signal Board
Later I changed my thinking and separated the fan control block, using a dedicated heavy copper PCB to carry it. I contacted several heavy copper pcb manufacturers and finally still chose a supplier that had made power electronics boards, with copper thickness directly up to 4 ounces, and although the board was as thick as a small brick, the stone in my heart landed when the connector terminals were soldered on. This board had no very high-density routing, and the signal part was made separately with a multilayer board and connected through pin headers, equivalent to a big one and a small two. Many friends doing high speed pcb manufacturer work may feel that the signal path becomes longer this way with a risk of cross-board reflection, but in fact for fan control, TACH is a low-speed pulse and PWM frequency is only tens of kilohertz, so there is no need to be that nervous. The effort saved from winding around for length matching is exchanged for power pins that do not heat however you tinker, and the cooling power supply of the whole cabinet is as steady as an old dog.
Questions I Ask a Supplier Now
Now when I choose a multilayer pcb supplier again, my first question is not how many layers they can make or how fine the line spacing can go, but whether they can bury copper blocks directly under the fan interface for heat dissipation, whether they can do local heavy copper, and whether the TACH signal lines have basic protective filtering. Because once there are many server fans, dozens of speed feedback signals squeezed together, ground bounce and crosstalk slowly eat your noise margin. A good board factory will, at the design stage, help you group the fan signals, and how to arrange the power and ground layers, and may even suggest you add an RC low pass at the TACH input, and these details are far stronger than simply chasing HDI any-layer interconnect.
Fault Prediction and Clean Waveforms
There is another place that is especially easy to overlook: whether fault prediction is truly reliable. Many people, upon getting a fan control board with speed monitoring, think everything is fine, set a lower speed limit and alarm below it. In real scenarios, from slight wear to a fully jammed bearing, there may be only a few hours in between, and very often the speed does not fall linearly but oscillates within some range like a seizure. If your controller catches only with a rigid threshold, it will probably either miss or falsely alarm and annoy people to death. Later I simply did a sliding window variance analysis of the TACH data through a small microcontroller, using speed stability to judge, and once the dispersion of speed under the same PWM suddenly grows, even if the absolute value has not fallen through the threshold, a warning is pushed in advance. This does not require upgrading hardware, and the key is that the TACH waveform on your PCB must be clean enough, so that normal jitter is not amplified into a fault signal.
Cost Comparison for Split Designs
Quite a few people ask me how much the price differs between a heavy copper pcb supplier and an ordinary hdi pcb manufacturer. Honestly, at small volume it is not much more expensive, and a control board used for a few more years, with one unexpected downtime loss, is enough to buy a box of boards. Moreover, once you split the fan board into a high-current backplane and a signal acquisition board, the signal part can even continue to use the HDI process and find an hdi pcb supplier to make fine lines, only leaving more margin for the power layer and ground layer. In this way, light and thin where it should be light and thin, and solid where it should be solid, with the same total cost, reliability instead rises by a notch.
More Posts
Battery Formation Tester PCB: Heavy Copper, Kelvin Sensing and Thermal Design for Accurate Multi-Channel Formation
Working on battery formation equipment, I learned that test accuracy depends not
Blood Glucose Meter PCB: Leakage Current, Impedance Control and Material Choices for Microamp Sensing
After years on blood glucose meter projects, I learned that a Blood
Su proveedor de confianza de fabricación de PCB y montaje de PCB todo en uno
- Experto en producción de lotes pequeños y medianos
- Fabricación de placas de circuito impreso de alta precisión y montaje automatizado
- Socio fiable para proyectos electrónicos OEM/ODM
Horario comercial: (de lunes a sábado) De 9:00 a 18:30