The 2321 Data Cell Drive was in full production at this time and many machines had been delivered. Therefore, each failure caused prolonged down time and possible data loss, which is a Cardinal Sin in the data storage business. Therefore, Dan Olsen and Stan Moss would have to work quickly as continued failures of the 2321 would reduce customer satisfaction and could be detrimental to future 2321 sales.

The IBM 2321 Data Cell Drive is a vast capacity, random access storage unit that stores data and retrieves data from magnetic strips (.005" x 2-1/4" x 13") under the direction of a storage control unit. Each strip is one of ten stored vertically in a sub-cell, with 20 sub-cells per cell and ten cells forming a circular array (Exhibit A-1). A strip is taken from the cell, wrapped around a drum (Exhibit A-2a) and passed by a read-write head. The 2321 is designed to be used with IBM computers and in its System/360 and 370 compatible form, it can store up to 400,000,000 8-bit bytes per array. The access time to a specific storage location varies from an average minimum of 175ms to an average maximum of 600 ms. These access times are remarkably fast considering the complexity of the mechanical accessing system. When the IBM 2841 Storage Control Unit specifies a storage location, the hydraulic servo of the 2321 rotates the circular array of storage cells until the proper subcell is located at the pickup point. Each strip has two tabs on its upper edge and a set of separation fingers isolates the desired strip at the pickup point (Exhibit A-2a). A latch keeper (Exhibit A-3c) then releases the torsion spring producing a high acceleration of the drum while a clutch engages the motor drive for continuous rotation (Exhibit A-3b, A-3a). As the drum rotates clockwise, the pickup head first moves down the chute to pick up the magnetic strip. The continuous rotation then pulls the pickup head and magnetic strip up the chute and on to the drum where the pickup head spring is locked in place (Exhibit A-2b, c, d). This strip is rotated past the read/write head and is then returned to its subcell. The return cycle starts with the compression of a torsional spring (similar to that shown in Exhibit A-3b) which then accelerates the drum counter clockwise as the clutch mechanism reverses the motor drive. The strip and head move down the chute guide, the strip is deposited in its subcell and the pickup head is returned to its ready position (Exhibit A-2d, c, b, a). The return to the ready position compresses the torsional spring (Exhibit A-3b) and sets the latch keeper (Exhibit A-3c) as a trigger.

The pickup head link shown in Exhibit A-2 was the result of three redesigns. The first two designs utilized a locking mechanism on the pickup head to hold the head to the drum during the read/write operation. The following design utilized a locking rod which rotated into place as the pickup head reached the drum (Exhibit A-2). The hook on this arm was later replaced with a spring and the final link and spring designs are shown in Exhibit A-4-1, Exhibit A-4-2, and Exhibit A-4-3.

The manufacture of the pickup head link started with an AISI 8620 steel forging. After both its rough machining and finish machining, the link had to be straightened as it become slightly twisted due to the releasing of internal stresses. Once the proper dimensions were attained, all but the tip area was plated with copper. The link was put into a carburizing oven, heated and quenched, carburizing the uncoated tip to produce a wear surface. The copper is then chemically removed from the link and replaced with a chrome plating for corrosion resistance.

When failures were first observed (see Exhibit A-4-1, assembly drawing), the links were taken to IBM's metallurgy group for analysis. Metallurgy reported that some surface decarburization was observed in the area of fracture. A decarburization indicates that carbon has gone out of solution, thus softening the metal. The link design engineer decided that a carbon enrichment process would bring the link surface back to its original carbon content and strength and eliminate the failures. With this change, new links went into production and testing. However, just after these new links were released they began to fail.

With the preceding background, Dan and Stan decided to start their investigation by recalculating the expected stresses at the link break point. They considered three loading sources; the spring load, the inertia load and the shock load. The spring load is generated when the link spring is locked to the drum. The calculation of this force existed in a previous 2321 report and Dan and Stan were satisfied with the 3. 8 lb. maximum force listed. A report also existed comparing theoretical inertial forces with experimental results. The theoretical results predicted a maximum acceleration of 54.2 g's on the 12.4-gram head and 3.0-gram magnetic strip. Experiments were run with a 30-inch mylar strip instrumented first with strain gauges and then with an accelerometer. The signals identified one acceleration peak as the head picked up the mylar strip and started back up the guide and another as the head hit the drum. Only the first peak is reflected as a load on the links and it was found to be 52.8 g's and 29 g's in the two experiments. Dan and Stan decided to use 54.2 g's as their design acceleration.

The final load source was shock loading at the cycle start and at latch. At the cycle start the high clockwise acceleration of the drum slams the pickup head against the outside of the guides. The friction between the pickup head shoes and the guides produces an impact force on the link. At the end of the cycle the pickup head deposits a strip into its subcell, and returns to the ready position (Exhibit A-2b to A2a). During its upward motion, the link is at a severe angle with the guides and friction between the pickup head shoes and the guides will load the links. However, a more severe load occurs at latch. As the drum reaches the end of its travel, the index roller rotates the lever compressing the torsion spring (Exhibit A-3b). At the latch (Exhibit A-3c) the latch striker moves past the spring-loaded latch keeper and impacts the latch bumper (the bumper is a safety over-ride feature). The drive shaft then reverses direction and is restrained by the latch keeper. However, the overshoot during bumper compression causes the pickup head to move further up and out against the guide. As the latch striker rebounds from the bumper, the pickup head shoes are against the outside of the guide and absorb the rebound forces. Occasionally, the link-guide angle exceeds the friction angle and the pickup head shoes lock in the guides, resulting in a system malfunction. It is almost impossible to accurately calculate these shock loads and experimental values have to be used in stress calculations.

Dan and Stan checked with the link designer and found that he was using the same spring and inertia forces that they had accepted and he was using a shock load of 15 lbs. The designer had taken this shock load from a single experimental result whose source had become somewhat obscure by this time. Stan had been involved in the original design of the 2321 links and his rather rough experiments had measured shock loads as high as 30 lbs. Many changes had occurred in the 2321 since Stan's experiments but he felt that 15 lbs was too low and that new experiments should be run to determine the shock loading in the present system. Dan assumed the responsibility for these experiments while Stan calculated the stress in the links using his old data of 30 lbs shock loading per link. Before Stan could start his stress calculations he had to determine the eccentricity at which the load was applied to the link. The eccentricity existed because all of the loading was transmitted through the link pin. By observing wear marks on the pins of broken links, Stan decided to use a loading eccentricity of .125 inch between the load point and the center of the link's .062-inch depth.