SPC charts are monitoring tools. They answer a specific question: has the process already changed from its established state? The Shewhart logic is elegant: calculate the natural variation in your process under stable conditions, define control limits at plus or minus three sigma, and flag observations that exceed those limits as signals of special cause variation worth investigating.
This is genuinely useful. It catches tool wear events, coolant contamination, fixturing drift, and operator-induced variation before they produce a full batch of out-of-spec parts. Shops that run SPC consistently produce better conformance than those that do not. That is well established.
The problem is scope. SPC answers the monitoring question, not the optimization question. Those two questions are different, and conflating them leads to a particular kind of suboptimality that is invisible to anyone looking only at control charts.
The Question SPC Does Not Ask
An X-bar and R chart for a bore diameter feature tells you whether the current process is in control relative to its own historical baseline. It does not tell you whether the current baseline is optimal. A process can be in statistical control and simultaneously running 5% below its theoretical yield ceiling because nobody has explored whether the current feed rate and spindle speed combination is the best available for this material and geometry combination.
Put another way: SPC monitors the stability of the current operating point. Parameter optimization asks whether the current operating point is worth being stable at.
These are not competing activities. They address different phases of the process lifecycle. The distinction matters because shops sometimes satisfy themselves with good SPC performance when the larger opportunity is that the operating point itself has never been fully optimized. A process with Cpk of 1.45 that is in control is a well-behaved process. Whether it could be a Cpk 1.75 process with different parameter settings is a different question that SPC cannot answer.
What Drift Patterns in SPC Actually Signal
When you do see meaningful patterns in SPC data, they are often more informative than the simple "in control vs. out of control" framing suggests. Consider a few common drift patterns and what they actually indicate about the parameter space:
Gradual trend toward one control limit: In a turning operation on hardened 4340 steel, a gradual upward trend on a surface roughness chart over 50 parts typically indicates tool wear. That is the standard interpretation. The less common follow-on question is: at what point in the tool wear curve is the process still capable of meeting spec? Shops often have a replacement interval that is based on historical experience. Whether that interval is optimal depends on what the parameter space looks like near the end of tool life: could a slight reduction in cutting speed extend the tool life window without degrading surface finish? SPC flags the trend; answering whether the interval could be extended requires parameter knowledge.
Sudden shift to a new stable level: A step change in the process mean, sustained over multiple subgroups, typically indicates a process input change: material lot change, operator change, coolant concentration change. Identifying the input change is the first diagnostic step. The follow-on question is whether the new stable level is acceptable and whether the parameters should be re-optimized for the new input conditions. A process that was optimized for one coolant concentration is not automatically optimized for a 15% change in that concentration, even if the shift is within spec.
High within-subgroup variation: When R-chart values are consistently elevated even when the X-bar chart is in control, the signal is not process drift but parameter-driven instability. High within-subgroup variation often points to chatter, thermal instability in the workpiece, or inconsistent chip breaking. These are parameter phenomena, not measurement phenomena. SPC identifies them, but the fix requires parameter investigation.
The Feedback Direction Problem
There is a directional mismatch between how SPC works and what process engineers need for optimization. SPC is backward-looking: it identifies that something has changed relative to a historical baseline. Parameter optimization is forward-looking: it asks what the process should be doing.
In practice, this means SPC data is underused as optimization input. The control chart data contains rich information about process behavior across a range of operating conditions, including conditions that arose from variation in inputs. A drill-down into the raw data behind a Cpk calculation often reveals that certain production windows, corresponding to specific operator shifts, coolant batches, or ambient temperature ranges, produced consistently higher yields than others. That variance is signal. But standard SPC analysis aggregates it away in the interest of process stability tracking.
We built the data integration layer in Reaxiomatic specifically to preserve this kind of sub-aggregate information. The parameter search engine needs to know not just what the current operating point is, but what outcomes were observed across the range of conditions the process has actually experienced. That is different from what a typical SPC export contains.
Where Parameter Optimization Picks Up
The starting question for parameter optimization is: given all the observations this process has produced, what combination of input parameters is associated with the best output outcomes? This question requires a different data structure than SPC monitoring.
SPC works on output measurements time-series: a feature dimension, a surface roughness value, a weight. It does not, in most implementations, store the specific parameter settings in effect at the time of each observation. It assumes the process is running at a fixed operating point and tracks variation around that point.
Parameter optimization requires the opposite structure: each observation needs to carry its input conditions (the parameter settings in effect) alongside the output measurement. Without that linkage, you cannot build a model of how inputs drive outputs. You can only know that outputs varied, not why.
This is the gap that most shops are sitting in. They have excellent output monitoring via SPC. They have incomplete or fragmented input documentation. Connecting those two data streams is the prerequisite for moving from monitoring to optimization.
Running Both, Not Choosing Between Them
The framing that SPC and parameter optimization are alternatives is worth rejecting explicitly. SPC belongs in production monitoring as a permanent feature. It catches real problems that have nothing to do with parameter settings: tool breakage, material deviation, machine fault states. No amount of parameter optimization eliminates the need to monitor whether the process is behaving as expected during production.
What changes with systematic parameter optimization is what you do when SPC signals something. Instead of simply restoring the process to its prior state after an out-of-control event, you ask whether the prior state was the right target. An SPC alarm triggered by a new material lot is an opportunity to re-evaluate whether the current parameters are still appropriate for the new lot's mechanical properties. That is a parameter optimization question, not a monitoring question, but it is prompted by the monitoring system.
The two disciplines reinforce each other when the data infrastructure connects them. That connection is harder to build than it should be in most shops, which is partly why so many organizations run excellent SPC programs and still have wide unexplored parameter space. The tools exist. The question is whether the data flows in a way that lets them talk to each other.