Maskininlärningsmodeller misslyckas inte på samma sätt som traditionell mjukvara gör. De kastar inte undantag eller kraschar. De fortsätter att producera förutsägelser, och förutsägelserna fortsätter att se rimliga ut, medan den bakomliggande noggrannheten tyst försämras. En McKinsey-granskning av AI-riskhantering noterar att modellövervakning i produktion behöver spåradata drift, model metrics, and bias metrics in outcomes as a continuous discipline rather than a one-time validation step.
That is a control problem, and it is exactly the kind of problem Lean Six Sigma practitioners have been solving with styrkort for decades.
Dataavvikelse och konceptavvikelse är två nya källor till speciella orsaker till variation
In a manufacturing process, variation comes from machines, materials, methods, and people. In a deployed model, variation comes from the data feeding it.
A quality manager doing rotorsaksanalys on a degraded model needs to separate these input-side shifts (data drift) from cases where the relationship between inputs and the correct output itself has changed (concept drift). Both are process variation. Neither shows up in a code review.
Mekaniken: centrumlinje, styrgränser och Western Electric-reglerna
A control chart applied to a model metric works the same way it works on a production line.
Applied to a model’s daily accuracy or error rate, these rules catch gradual degradation that a simple accuracy threshold would miss.
Ett kliniskt fall värt att studera
A 2024 peer-reviewed framework built for radiological AI systems shows what this looks like in practice. Researchers combined SPC with machine learning to catch
That is an average-run-length problem, the same metric used to judge how quickly a traditional control chart flags a shifted process.
Mjukvarubaserade diagram förstärkta med maskininlärning börjar ersätta enkla tröskelvärden
SPC for ML drift is not staying static either.
, judging performance by average run length. This matters for practitioners choosing between a standard Processduglighet approach and something more adaptive, an exercise no different from the tool-selection decisions taught in any Lean Six Sigma tool selectionmodul. En nyare rapport från 2025 om bootstrappade styrgränser för konceptavvikelse sträcker sig ännu längre och bygger styrdiagram specifikt kring modellens förutsägelsepoäng snarare än rå särdragsstatistik, vilket undviker problemet med falska larm som vanliga Sigma-tröskelvärden skapar när data inte är normalfördelade.
Att passa in modellövervakning i en DMAIC-styrplan
None of this replaces the discipline of DMAIC. Ett modellövervakningsprogram gör sig förtjänt av sin plats i Control-fasen på samma sätt som vilken stabiliserad process som helst: definiera mätvärdet (noggrannhet, felprocent, ett särdrags medelvärde eller proportion), sätt gränser utifrån en stabil baslinjeperiod och tilldela ägarskap för att utreda överträdelser. Databricks ingengörsteam uttrycker det kort och gott när de beskriver ML-pipeliner i produktion:
, which is a governance description of exactly what a Black Belt would call moving from measurement to control.
Vem äger diagrammet när modellen har levererats
Someone has to own the breach investigation, and that ownership question is where most model monitoring programs quietly fail. A Black Belt already knows how to write an escalation path for an out-of-control signal on a machined part; the same skill transfers to a model accuracy chart, provided the data science team documents baseline behavior clearly enough for someone outside the model-building team to read it. Leadership sponsorship matters here specifically because model drift alerts are easy to dismiss as noise until a credit, fraud, or clinical decision goes wrong in production. Building that ownership into a project charter, with a named process owner and a documented response threshold, is standard DMAIC practice applied to a newer kind of process.
Källor
- Databricks (2019), Productionizing Machine Learning: From Deployment to Drift Detection: https://www.databricks.com/blog/2019/09/18/productionizing-machine-learning-from-deployment-to-drift-detection.html
- Zamzmi, G. et al. (2024/2025), Out-of-Distribution Detection and Radiological Data Monitoring Using Statistical Process Control, PMC: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11950585/
- arXiv (2024), Out-of-Distribution Detection and Data Drift Monitoring using Statistical Process Control: https://arxiv.org/abs/2402.08088
- arXiv (2025), A Review of Artificial Intelligence Impacting Statistical Process Monitoring and Future Directions: https://arxiv.org/pdf/2503.01858
- arXiv (2025), Bootstrapped Control Limits for Score-Based Concept Drift Control Charts: https://arxiv.org/pdf/2507.16749
- JANUS Associates, NIST.AI.100-1 AI Risk Management Framework overview: https://janusassociates.com/resources/nist-ai-risk-management-framework/
