Hot-Swap Imager Adapter

Client

Location

Pittsburgh, PA

Year

2022

During my Summer of 2022 internship at Carnegie Robotics, LLC, I was tasked with designing a way to allow imager PCBs housing camera lenses to be "hot-swapped" during calibration and assembly, meaning that they could be disconnected from the text fixture without damaging the delicate circuitry inside.

I iterated through many solutions before I successfully designed and fabricated a PCB capable of functioning exactly to spec, an ESD-safe mechanical housing, and, most importantly, an intuitive Tangible User Interface (TUI.) This iteration process was greatly accelerated by my development of an in-house PCB prototyping process, involving milling 2-sided boards in minutes.

Problem Statement

Carnegie Robotics makes a set of MultiSense Stereo Cameras for a wide variety of applications that require reliable, robust equipment such as mining, construction, off-road navigation, and more. Each camera module has a set of camera lenses through which real space can be mapped to camera space for DSP purposes.

Each of these camera lenses is attached to a small imager PCB that both draws power from and transmits/receives data from the main Power/IO board at the core of the camera module. Each of these imager board lenses must be precisely focused and calibrated before the final installation in the stereo camera module, and we do this through a process represented by the abstract diagram below:

Connecting/disconnecting imager boards directly to the connecting cable when the processor board is powered on risks damaging delicate circuitry in the imager board, so the processor board must be depowered whenever an imager board is being connected/disconnected before calibration can begin. The processor board's firmware takes a few minutes to initialize every time it is powered up, so this boot-up delay during the process of focusing a large set of imager boards sums up to a massive amount of wasted time.

With this inefficiency hampering the productivity of assembly technicians, I was tasked with finding a solution. So, I asked myself:

How might we create a solution to the delay induced by power cycling the stereo camera test fixture every time we want to focus a new imager?

Solution & TUI Design

While the cables connecting the power board to the imager boards house quite a large array of parallel wires, only some of the wires pose a legitimate threat to the imager boards' circuitry. Each wire in the cable serves a different purpose - some are purely power lines, some are grounded, and the rest are data lines, either transferring data to or from the imager boards.

If I could effectively "cut off" both the voltage lines that deliver power to the imager board and the data lines that the power board uses to command the imager board, the imager board would not be sending signals back to the power board, and it would be safe to disconnect the cable. Abstractly, my solution would look as such:

I began, of course, with the end user in mind - the assembly technician. Specifically, what is a tangible user interface (TUI) that will allow assembly technicians to safely and rapidly cut off dangerous cable lines, swap out imager boards, and then re-enable the cable lines so that the new imager can be focused?

I decided that a simple 3-position rotary selector switch (3PST) would be the simplest TUI to use. While not nearly as easy to find as 3PST toggle switches or DPST slide switches, the user-centric act of torquing a knob to different positions with a supporting visual interface of text and LEDs was sufficiently important enough to me to justify the supply-chain-nightmare challenge.

I breadboarded the interface to determine the switch's operation and determine what would make sense for the TUI. I settled on the three positions of the switch corresponding to the following actions:

"OFF - Manual" - The center position, indicated by a red LED, turns the dangerous cable lines OFF. The imager can now be safely removed.
"ON - Manual" - Rotating the switch to the right position will turn the cable lines ON, indicated by a green LED. The imager is able to be focused, but the dangerous cable lines will have to be turned OFF before the imager can be safely removed.
"AUTO - processor control" - while not yet developed, I added the functionality of a processor being able to connect to my solution and control the switch with electrical signals, rather than a manual switch throw. This state is indicated by an orange LED.

I laid out the TUI in CAD as shown to the left, with the 3 LEDs indicating the 3 states described above. I also included a power indicator LED.

Circuit Design

I chose to use 3 different circuits to act as switches for the dangerous cable lines and routed each line into its respective circuit depending on its nature.

For the power lines, I used a load switch IC, as they would be powering both the imager board and my PCB. I chose the IC to accommodate the current carried by the lines and do suppress any switching transients that they might exhibit.
There are multiple digital communication protocols involved in the TX/RX between the imager and IO boards, two of which were used in processor --> imager inputs, and therefore dangerous.

For the protocol that is bi-directional, I used an array of relays, as they allow bi-directional data transfer.

For the protocol that is uni-directional and requires more pins, I used an array of tri-state buffers.

Both circuitry solutions for the data lines meet the requirements of carrying less current but requiring high switching speeds.

I breadboarded with each IC to test its functionality, including the use of SMT-to-DIP adapters for the surface mount components.

Below is a system diagram of the PCB design. As can be seen, the lines in the connector cable are split up and sent into the corresponding circuitry depending on its type. The circuitry for the dangerous lines (power, data) are controlled by a single digital enable signal, which can be controlled manually via a rotary switch or by a processor.

PCB Prototyping

Thanks to the machinery in CRL's machine shop, I was able to develop a system that allows CRL engineers to mill 2-sided PCBs in a matter of hours, as opposed to the weeks that it usually takes for professionally-made PCBs to arrive. This is an incredible capability that allows iteration cycles of simple circuits, adapter boards, and board cutouts to speed up tremendously.

The Hot-Swap Imager Adapter PCB is an excellent example of this in-house prototyping capability. I ordered my first PCB iteration to be produced professionally, as seen with the red silk screen below. You'll notice many errors: un-coupled and non-length-matched differential pair sets, lack of stitching vias on high-speed data lines, overly-long trace lengths, and a less-than-optimal ground plane on the bottom.

While I waited weeks for the simple PCB to be fabricated & shipped, I worked to mill my board in-house in parallel, a bare version of which is shown on the left. I went through 13 iterations, each containing many ripped & lifted traces, rough cut lines, and much solder paste.

I plated the vias by passing a bare wire through the milled holes in the board and soldering both sides. This introduced me to the constraint of not being able to place vias under components, as manually-filled vias bulge out of the board.

I still milled a working board in-house well before my professionally made PCB arrived. By that point, I'd made so many design improvements that soldering the professional PCB wasn't worth my time.

You'll see an ugly red wire connecting two points, clearly indicative of a late-stage design change. That is the beauty of prototyping with bare boards - changes like these can be easily made and tested well before the final fabrication stage.

Enclosure Design

I designed an enclosure that was 3D-printed with ESD-safe material. This housing must protect the delicate circuitry inside while allowing easy access to PCB connectors and test points. It also needs to display a simple and effective user interface. Below shows an animated exploded view of my solution.

The housing encloses the PCB whilst leaving off-board connectors and test points exposed for easy access. I also incorporated a cable strain relief solution that integrates into the bottom 3D printed base to robustify the design for the calibration room.

The rotary selector has panel mount capability with a threaded shaft. The LEDs are pressed into the top panel through plastic beveled LED holders. The base and top panel are fastened together via machine screws and heat-set threaded inserts. The PCB also mounts to the base through machine screws and has additional support from standoffs.

Final PCB & Enclosure assembly

Thanks to my in-house milling process, I could guarantee that my PCB was fully functional. With this in mind, I transferred my board to a 4-layer layout and ordered a professional fabrication job.

I assembled the TUI, including crimped & connected JST connectors. The full assembly with labels is shown below on the right. The final unit works exactly as intended. The next steps for CRL are to duplicate multiple versions of this unit so assembly techs can work in parallel and develop a protocol for automated processor focusing.