System chassis design follies

Designers can approach building a system chassis to meet design specs in several ways. Some of them result in a 'help-me box' — a chassis that begs for improvement to meet the full intent of a spec. The 'spec intent' approach avoids this by looking for system improvement opportunities and creating value-added designs, while controlling costs at the same time.

Oct 1st, 2003

By Matt Tracewell

Designers can approach building a system chassis to meet design specs in several ways. Some of them result in a 'help-me box' — a chassis that begs for improvement to meet the full intent of a spec. The 'spec intent' approach avoids this by looking for system improvement opportunities and creating value-added designs, while controlling costs at the same time.

A frequent folly in system integration is building to a design specification without adequate consideration of how individual components and subsystems work together. Simply meeting the spec and its basic functions does not necessarily achieve maximum functionality or reliability.

There may also be other detrimental results, such as poor thermal performance, electromagnetic interference shielding problems, and maintenance difficulties. Attaining the full intent of a chassis design spec requires an in-depth analysis of how individual components could be designed and assembled for the best results, and some innovative engineering to make it happen.

In the system-packaging industry there are many different product perspectives, design approaches, and competency levels. For example, the DIN rail 'erector-set' approach follows a much different path than the one taken by a metal fabricator or machine shop.

Both of these approaches differ from those of a contract house that designs everything from a blank sheet of paper, or an integrator that purchases and assembles catalog items. A 'spec intent' approach might use elements of these other design tactics, but by focusing skills and experience on the desired outcome, the specifier gets what he wants and needs. Simply building to spec by purchasing and assembling components on a bill of materials often results in a 'help-me box.'

Take for example a MIL-STD-461E VME system. A 'spec intent' approach can help the designer avoid a system chassis that begs for improvement. Recently a military/aerospace system supplier sent our requests for information on a system chassis with a requested delivery lead-time of 12 weeks. Specifications for this system are:

  • 17.5 inches high by 17 inches wide by 19.8 inches deep;
  • 19-slot 6U by 160-millimeter front-loading card cage;
  • custom 19-slot monolithic J1/J2 VME backplane per specification;
  • J1 with 96-pin DIN41612 short-tail connectors;
  • J2 with 16-pin VME64X long-tail connectors with screw-down shrouds;
  • active terminations and active daisy chain;
  • input of 32 volts DC to 56 volts DC;
  • output of 5 volts at 120 amps, plus-or-minus 12 volts at 12 amps;
  • redundant current-sharing power supplies
  • hot-swap current control;
  • power control per slot;
  • voltage monitoring and system status signaling;
  • cooling by two 375-cubic-feet-per-second fans;
  • air flowing to the chassis from the front panel to rear of the unit;
  • fan alarm provides signal upon failure;
  • air inlet and exhaust is shielded with honeycomb filters;
  • removable, cleanable air filter;
  • digital crosstalk isolation not less than 90 dB;
  • analog crosstalk not less than 100 dB;
  • shielding for electromagnetic and radio frequency interference (EMI and RFI) that meets MIL-STD-461E;
  • Tempest-compliant line filter; and
  • 19-inch rack-mountable with EIA flanges and chassis slides.

The supplier awarded the contract, and the winner proceeded to design the chassis using an 'erector-set' approach. Designers bought and assembled components, and tested the resulting chassis to verify compliance to the specification. The designer delivered the chassis to the customer within the scheduled lead-time.

With the VME cards in place, the customer started a series of system performance tests, and immediately uncovered several problems. These included various component limitations, assembly practices that contributed to a cluttered layout, lack of true hot-swap capability for the power supplies, inconvenient fuse and fan servicing, thermal hot spots, and a many access plates and fasteners that compromised EMI suppression.

Subsequently, Tracewell Systems Inc. in Westerville, Ohio, was asked to evaluate the design and suggest improvements. Using an engineering approach called 'spec intent' the firm was able to create a system chassis that offered many advantages in performance and serviceability, while maintaining a price similar to the more traditional solution. By viewing the functions of the chassis, backplane, power, cooling, and interconnect system as unified entirety, the spec-intent approach resulted in a tightly integrated system that allowed:

  • true hot-swappable redundant power from the front of the chassis;
  • elimination of backplane fuses and hot-swap coils;
  • remote reset power protection;
  • pluggability of the hot-swap circuitry/power protection on a per-slot card level;
  • fan replacement without disassembly of the chassis;
  • substantial improvement in airflow efficiency;
  • elimination of a power supply output/remote sense wiring harness; and
  • reduction in the number of access plates, screws, and other fasteners.

This and similar experiences helped identify several design follies, and ways in which the spec-intent approach can avoid the problems they create. Typically, most of these situations start out the same way: generate design goals, translate them into specs, request quotes, and award work contracts. Specific design areas for this example point out ways to improve on final results.

Ruggedized package

First, designers had to create a ruggedized package and provide access to internal components for repair and calibration. The design folly happened when engineers chose or designed a welded aluminum chassis to meet spec's size and ruggedization requirements; the designer had to add panels and hatches for access to internal equipment after internal layout.

Problems of this approach included five access doors or panels that increased reliability risks. The doors introduced a strong possibility of misaligned gaskets that might increase RFI. Gasketing of about 300 linear inches, a fastener count of about 150, and a high component count all added to cost and assembly/disassembly time. Many access doors or panels also were signs that other aspects of internal system design and layout might create additional problems.

Yet a spec-intent solution involved modifying the standard welded aluminum chassis by keeping the number of access doors and plates to a minimum and designing the internal layout for ease of maintenance. Accordingly, this approach improved the design by reducing the number of access panels from five to two, reducing the total length of gasketing from 300 to 112 linear inches, and reducing the number of captive fasteners from more than 150 to only 17.

Blending communications and power

Next, designers set a goal of placing communications and power on the backplane. The design folly of this was a custom backplane with integral hot-swap coils, indicator fuse, and switch. Power cables from diodes entered the backplane and routed through these components to each separate slot. Potential problems included complex power distribution and signal path, as well as six interconnection points between the power supply and backplane for main power input.

A spec-intent solution modified the standard VME backplane with custom signal routing and independent power routing to the slots. Designers built a separate power plane board to eliminate multipoint wire input and several different signal inputs.

The outputs of the power supply modules combined on the power plane into a bank of holdup/filter capacitors before distributing to 19 power control assemblies. All additional interconnect, signal routing, fan power, and monitoring resided on this board. The power plane attaches with a series of 60-amp 'power bolts' that act as a mechanical connection as well as low-resistance system ground path.

This approach introduced improvements such as providing a power plane architecture with a centralized, low-impedance ground path, as well as 73 feet less wiring, and 36 fewer power entry tabs.

Power input control

Another example revolves around a power input control in the chassis. The design folly involves using a single-pole 50-amp circuit breaker with a front-panel off/on switch. The problems this introduces include a high-current front panel switch and bulky high-current DC wiring, as well as an increased risk of EMI problems, thermal hot spots, and a single point of failure.

The spec-intent solution substituted a power-control assembly (PCA) for input power control. The PCA had four electronic switch and circuit breaker circuits, one for each VME system voltage and one for the distributed voltage. Controlling these breaker circuits is a gate signal from the backplane that indicates when a board is installed and ready to be powered up in a controlled way; each circuit is resettable with a switch for each slot or a single maser switch for all card slots.

The system-enable signal from the front-panel on/standby switch combines with these individual gate signals. The power line filter has two sets of output terminals. Essentially, it takes one input and splits it into two separately filtered outputs. One input is for the system power supplies, and the other is for pass-through voltage on the boards.

Improvements of this approach include electronic control of the pass-through voltage at each slot that enables the entire system to go into standby with a single-pole low current switch, replacing the single-pole 50-amp circuit breaker on the front panel with a two-pole 25-amp breaker on the rear channel, and improved protection level of the chassis.

Redundant power supply

Another piece of the design involves establishing a redundant power supply. The folly of this approach involved a standard Vicor MegaPac divided into side A and B for redundancy that included external diodes. Connectors in the output harness enabled supply replacement.

Problems revolved around a power supply that does not have a separate source of cooling air; it adds to the heat load of system fans. In addition, the supply is located in a difficult access point and its output cable has six connections between supply and backplane. If section A fails and runs on section B, users must take the chassis out of service and take it apart to replace the supply.

The spec-intent solution uses a modified standard power supply based on a high-density, high-reliability DC-DC converter. It consists of three identical power modules in an n+1 redundant configuration with integral power metal oxide silicon field effect transistor (MOSFET) OR-ing circuits. To meet total load specs, two supplies operate in parallel at all times with output per tray of plus-or-minus 5 volts at 60 amps and plus-or-minus 12 volts at 6.25 amps. Each redundant power module contains visual status indicators, remote status signals, precision output voltage monitors, enable/disable switch, as well as over voltage, over current, and short-circuit protection.

Power supply cooling air is isolated from the main system and passes through a bonded folded fin assembly with fin density that changes easily to match airflow and temperature requirements. Each power module is removable without the use of tools while the entire system remains in operation.

Improvements that this approach brings include front-accessible pluggable power modules; true hot swap replacement with no power down or excessive teardown; improved cooling that eliminates temperature-rise problems; substantially reduced wiring and connector components; and external OR-ing diodes and monitoring components eliminated.

In another part of the chassis, designers tried to maintain exhaust temperature at or cooler than 5 degrees Celsius above input temperature. A design folly related to this goal involved using dual fans with the air-input area — restricted to 12 square inches — far smaller than outflow area of 53 square inches. The fan location was difficult to access, and maintainers would have to remove more than 22 fasteners to replace the fan.

Problems included fans that were not truly hot swappable, replacement that requires off-line disassembly, and temperatures that might have been lower with better flow. The spec-intent solution improved airflow areas in inlet and outlet, and added hinged panels for fan access at the rear of the chassis. Fans are connected to the power plane.

In monitoring system voltages, the design folly involved a discrete centralized monitoring circuit board with I/O wiring mounted in the chassis. This created wiring clutter and reliability problems, a potential single point of monitoring failure, and a monitoring board that was not hot swappable.

The spec-intent solution used distributed monitoring within the power supply and fan assemblies for more reliable fault detection by removing the single point of failure, added hot swappable power supply and fan assemblies, simplified wiring, and reduced parts count.

Designers needed to establish EMI/RFI shielding to meet MIL-STD-461E and Tempest standards, so they originally cooled air ports with honeycomb filters, and added shielding gaskets to the five access doors on the chassis. but this added components and increased the number of potential failure points, as well as restricting cooling airflow with the honeycomb filters.

The spec-intent solution shielded fan inputs and outputs, leaving only two access doors necessary for servicing. Bent panel subassemblies and welded or riveted construction reduced the need for separate gaskets and shielding.

Designers also had to maintain low analog and digital crosstalk between cables. The original design placed wire and cables wherever convenient, leaving excessive components and poor component placement next to sensitive interconnects. The spec-intent solution eliminated cables wherever possible, and routed and isolated cables to control cross talk.

Matt Tracewell is executive vice president of Tracewell Systems Inc. in Westerville, Ohio. His e-mail address is mtracewell@tracewell.com. For more information contact the company of the World Wide Web at http://www.tracewellsystems.com/.

More in Power