3 Knowledge in a dynamic world.

The treatment of knowledge changed fundamentally in the dynamic world that followed WW II. Problem solving and learning, which entailed the development of new knowledge, had to become organic to the production process. Finding a single optimum production method was replaced by change as the central concern of manufacturing.

The first three epochs emphasized increasing mechanization in a

world that was, at least ideally, static--doing the same tasks again

and again, as efficiently as possible, at increasingly high volume.

Discretion was progressively removed from workers, and knowledge

about their tasks was subdivided and given to specialists, removing

it from the shop floor ... In contrast, in the last three epochs,

while the tools continued to become more mechanized, knowledge about

the work was returned to workers and their discretion increased.

The key goal shifted from efficiency at high volume to coping with

a dynamic world of rapid changes such as high product variety

and rapid product introduction. [15, Section 6]

But mechanization, in particular the development of increasingly autonomous machines, continued unabated. For a machine to operate autonomously a high level of knowledge is needed to guide responses to or forestall disruptions. Taylor's contributions to knowledge management discussed in the previous section thus continued to be vital, even as his approach to shop floor management was being turned on its head.

The Statistical Process Control (SPC) epoch coincided with a flowering of academic research on the science of metal cutting. Taylor's attempt to determine empirical formulas for factors that affect the rate of machining was extended, with the goal of raising effective machine speeds and productivity through a deeper understanding of the underlying science. This research was not primarily concerned with the SPC agenda of controlling variation.

The organization of this Section is not strictly chronological. We first examine the knowledge effects of the SPC epoch and the coinciding academic development of the "engineering science of machining." We then explore how numerical control initially foundered for want of sufficient knowledge. Finally, we consider what happens with fundamentally different manufacturing processes.

3.1. Statistical Process Control Epoch

The SPC epoch arrived at Beretta in the 1950s with the contract to manufacture the Garand M1 rifle. [15, Section 6] SPC shifted concern from average performance to variations in performance. To understand causes of variation requires detailed knowledge about a process and its real-world operation. Beretta's newly formed quality control department "was responsible for quantitatively measuring the natural variability of every machine and the degree of fidelity of every tool, verifying tool conformity to design, and identifying possible causes of systematic error."

Because so many variables can disturb a process, the complexity of causal networks for variation is an order of magnitude higher than for ideal operation. SPC thus drove the development of much more detailed causal knowledge, with a strong emphasis on the actual behavior of processes and machines on the factory floor.

This reorientation was accompanied by a complementary shift from a static to a dynamic world view. Dynamic causal models, in which sources and consequences of changes are explicitly monitored over time are vital to SPC. Each variable becomes a time series. Dynamic behavior such as the rate at which variables change had to go from being recognized to being measured (via control charts) to being adjustable. To eliminate adjustments between setups, for example, the rate of drift of key variables had to be constrained. But because dynamic behavior in this period was still not technically capable, processes escaped from control and interventions continued to be necessary. (1)

"Soft" innovations, such as control charts, were a hallmark of this period. The genius of the control chart is that it enabled operators, in a pre-computer era, to track dynamic variables and filter out real shifts from normal stochastic variation. Beretta's quality control department employed a variety of even more sophisticated statistical techniques such as gauge R&R studies, which are still essential for physical measurement. (2)

These changes shifted the focus of manufacturing from control to learning. "The application of SPC provided one way by which errors could, over time, be observed, better understood, and eventually solved. Manufacturing's evolution from an art to a science now included a systematic way of learning by doing." They also directed attention away from the product to the process. SPC effectively democratized and replicated Taylor's innovations in systematic learning about processes, even as his de-skilling of line workers was being reversed. Modern versions of SPC, such as Total Quality Management and Six Sigma, have institutionalized systematic learning, and moved it from the factory floor into general management.

Beretta's introduction of synchronous lines both required and made easier an integrated view of production, involving analysis of interactions among variables in different parts of a process. The sequence of workstations that comprise a process could no longer be assumed to be independent. This necessitated a major shift in problem solving and learning from a focus on individual machine performance to a process orientation. (3) "Diagnosis and problem solving are now carried out by examining the workstation not in isolation, but as part of the entire system ... Synchronous lines forced an integrated view of the entire system of manufacture. Whereas the intellectual underpinnings of Taylorism were reductionism and specialization, that of SPC in a synchronous line was integration."

3.2. The Science of Cutting Metal

At roughly the same time that Beretta was introducing SPC, formal laboratory-based research into machining was being conducted by universities and company research labs. Much of this research emphasized machining-speed issues in the Taylorist tradition, over precision and quality which are central concerns of SPC. A distinguishing feature was the effort to develop models based on known scientific principles rather than just fit curves to empirical data.

The basic characteristic of science-based modeling of machining

is that it draws on the established natural sciences, and

particularly the science of physics, to establish reliable

predictive models. These are models that can then be used to carry

out reliable engineering calculations of the expected behavior or

characteristics of a machining process, independent of empirical

information.

Development of capability for science-based modeling of

machining was quite dependent on the knowledge and understanding

of machining developed by the [earlier] research on empirical

modeling. A good example of such was the research done by the

Ernst-Merchant team ... in the period from 1936 to 1957, which

culminated in the creation by Merchant of the basic science-based

model of the machining process. [26]

Researchers found, for example, that the shear angle, the angle at which metal chips "peel away" from the face being machined, was key to predicting machining behavior. Shear angle being an important intermediate variable, it became a target for detailed causal modeling. "The ultimate goal of the above analysis leading to the shear angle relationships is to enable the estimation of all the relevant metal cutting quantities of interest, such as the forces, stresses, strains, strain rates, velocities, and energies without actually measuring them. For example ... knowing the shear stress of the metal and the cutting conditions, all of the above metal cutting quantities of interest can be calculated." [19, p 86, emphasis added] We can thus say that for the first time the knowledge graph incorporated "first principle" scientific models.

Among its major accomplishments this research: (4)

* Extended Taylor's empirical research to a range of additional operations (turning, milling, drilling) and issues (surface finish, costs, forces);

* Established a qualitative understanding of what happens when a tool cuts. The research identified four basic processes: primary shear, secondary shear, fracture, and built-up edge formation. These correspond to four distinct causal models with only modest overlap; [26]

* Yielded further details of cutting tool design, including materials and geometries for different purposes;

* Originated theoretically based models of the forces at work in metal cutting (e.g., Figure 3.1);

[FIGURE 3.1 OMITTED]

* Contributed analytic models of heat and thermal effects in metal processing.

In addition to incorporating fundamental scientific models for the first time, this research was notable for its depth. More variables and more relationships were incorporated into knowledge graphs, reflecting the fractal nature of causal knowledge. The more closely a phenomenon is examined, the more complex it appears. The effects include:

* Individual variables are replaced by collections of more specific variables.

* When a variable is discovered to be important, its causes must be understood in turn.

* New relationships among variables are identified, so a causal knowledge subgraph that is initially tree-like becomes a more complex network.