Process Monitoring — SPC, Multivariate Analysis & Fault Detection
Detect process drift, abnormal events, and incipient failures before they become quality exceedances or unplanned shutdowns — using the data your DCS already collects.
Book a 30-min technical fit callPhD APC/MPC · CERN · IAV · SkySails
What Process Monitoring Does
Statistical process control (SPC) and multivariate process monitoring detect when a process leaves its normal operating envelope — before the deviation causes quality problems, waste or equipment damage.
For simple single-variable processes, univariate Shewhart charts detect shifts in mean and variance. For processes with hundreds of correlated sensors, multivariate SPC (MSPC) with principal component analysis (PCA) and partial least squares (PLS) monitors the correlation structure itself — and detects subtle drifts that single-variable charts miss.
Process monitoring is the diagnostic foundation. It does not control the process; it tells you that something is wrong, where the deviation started and how severe it is. That makes it the natural prerequisite for APC: before closing the loop with a model-based controller, you need to know the process runs normally.
Three Monitoring Methods
Univariate SPC — Shewhart, CUSUM, EWMA
Tracks individual variables against control limits derived from historical data. Shewhart charts catch sudden shifts, CUSUM gradual trends, EWMA weights recent observations more. Standard in discrete manufacturing and batch release. Limitation: it misses correlated multi-variable drifts.
Multivariate SPC (MSPC) with PCA/PLS
Projects many correlated process variables into a low-dimensional latent space and monitors two statistics: Hotelling's T² (variation inside the model) and SPE/Q (residuals outside it). A change in either signals abnormal operation — even when no single variable exceeds its limit. Used in continuous chemical processes, semiconductor fabs and pharmaceutical batch control.
Fault detection and diagnosis / anomaly detection
Data-driven anomaly scores from isolation forests, autoencoders or Gaussian-process models; contribution plots show which variables drive the deviation. Used for early equipment-fault detection and as precursors of process upsets. Extends MSPC with diagnostic capability.
Where It Is Applied
Continuous chemical processes
Monitor reactor temperature profiles, column compositions and recycle loops together. Detect declining catalyst activity, heat-exchanger fouling and feed-quality excursions before they cause off-spec product.
Pharmaceutical batch manufacturing (batch MSPC)
Compare each batch trajectory with a golden-batch reference and detect deviations while the batch is still running, so corrective action is possible before the batch fails. The FDA PAT guidance supports this approach.
Semiconductor fault detection and classification (FDC)
Tool-state monitoring for etch, deposition, CMP and lithography chambers: tool drift, chamber-wall deposits and recipe deviations detected from tool telemetry — before wafer measurements confirm out-of-spec product.
Process health check before APC
Before deploying APC or MPC, MSPC shows which variables are in statistical control and which show special-cause variation. Controllers deployed on poorly understood processes fail quickly; monitoring first makes the control foundation solid.
How We Implement
Historical data analysis
Analyse historian data (e.g. AVEVA PI, Aspen IP.21, Ignition) for correlations, steady-state periods, grade transitions and outliers, and establish the normal operating envelope.
PCA/PLS model development
Build latent-variable models of the dominant correlations, choose the monitoring statistics, set limits for the acceptable false-alarm rate and validate them against known fault events in the history.
Contribution analysis and alarm design
Configure contribution plots that point operators to the most influential variables during an alarm, and set thresholds that keep false alarms low without losing sensitivity.
Integration with DCS and historian
Deploy the monitoring models as scheduled historian calculations or real-time streaming applications, with operator screens for trends, T²/SPE charts and alarm summaries.
What You Gain
Detects what single alarms miss
A process can drift a long way before any single variable reaches its alarm limit. MSPC detects the change in the pattern, not just individual exceedances.
Fewer unplanned shutdowns
Early detection of precursors — equipment degradation, feed drift, utility fluctuations — allows preventive action in planned windows instead of emergency response.
A prerequisite for APC, not a replacement
Continuous monitoring confirms that the APC/MPC model is still valid. When monitoring detects abnormal operation, APC can switch to advisory mode — protecting both product quality and the controller.
Faster root-cause analysis
Contribution plots point to the variables behind a deviation, so engineers start the investigation in the right place.
Where It Fits in the Stack
Process monitoring sits at the base of the optimization stack. It needs no APC or soft sensors — it works on the DCS/historian data you already have. But it enables everything above it: APC should not be deployed on a process that is not in statistical control, and soft sensors should be validated against MSPC before they become APC inputs.
Relevant Design Patterns
Book a 30-min technical fit call
If you want to establish statistical process monitoring, validate an existing alarm system, or lay the foundation for APC deployment — a 30-minute call is enough to scope the work.
Book a 30-min technical fit call →