Statistical Techniques for Monitoring Industrial Processes

Explore Python, a high level, versatile language for simple calculations and data science workflows, backed by a vast community and packages for data collection, processing, visualization, and modeling.

Learn Python language basics, including data types (integers, floats, strings, booleans), lists, range, list comprehension, indexing, slicing, and core control structures such as for loops and functions.

Learn numpy for scientific numerical computing, creating one- and two-dimensional arrays, using arange, zeros, and identity matrix, and applying stacking, slicing, computing sums, and safe copies.

Pandas enables data manipulation and analysis with label-based slicing, SQL-like filtering, and time series features, teaching how to create Series and DataFrames and perform grouping and aggregation.

Explores popular Python packages for industrial data analysis, including matplotlib, seaborn, SciPy, statsmodels, scikit-learn, and Keras, with examples of plotting, interpolation, and time-series analysis.

Explore univariate monitoring with control charts to track a single process signal, using center line, upper and lower control limits, and three-sigma thresholds for in-control decisions.

Explore Shewhart control charts for monitoring process variables, compare three-sigma and two-sigma limits, understand false alert rates and sensitivity to drift, and estimate mean and standard deviation from historical data.

Explore how Cusum control charts detect small mean drifts in industrial processes, contrast them with Shewhart charts, and interpret two curves, thresholds, and drift parameters.

Learn how to apply a custom Cusum control chart to monitor the air flow rate into an aeration tank, detect upward drift, and evaluate performance on real industrial data.

Quantify control chart performance by balancing false alarm rate, missed detection rate, and fault detection rate, compare Shewhart and Cusum charts, and use RL to assess detection speed.

Explore the exponentially weighted moving average (EWMA) control chart, using lambda to weight past observations, with lambda-dependent three-sigma limits, illustrated on an aeration tank case study.

Explore the limitations of univariate control charts for multivariate, correlated data, and how PCA, ICA, and residual-based charts improve monitoring.

Explore multivariate spm for plant wide and equipment level monitoring, leveraging PCA and Pls to handle steady state, dynamic, and nonlinear processes for fault detection and diagnosis.

Explore principal component analysis as a dimensionality reduction tool in Python with sklearn, showing data scaling, fitting, transforming to one dimension, and reconstructing for a process monitoring solution.

Explore the inner workings of PCA, including loading vectors, score matrices, and eigenvalue decomposition, to understand how principal components capture variance and guide dimensionality reduction.

Apply principal component analysis to an industrial polymer manufacturing dataset to reduce dimensionality and remove correlations. Explore how many components explain 90% of the variance.

Explore PCA-based fault detection by computing the T-squared and Q (SP) metrics to separate systematic variation from noise, and use control charts with empirical or statistical upper limits.

Apply PCA to polymer manufacturing data to detect faults by computing SP and t square metrics, setting control limits, and evaluating Q and t square on training and test samples.

Use principal component analysis for fault diagnosis and fault isolation by analyzing contributions to t-squared and q metrics. Build contribution plots to identify the variables most responsible for process faults.

Explore fault isolation with PCA by computing ESP and t squared contributions on a polymer manufacturing data set, visualize the contributions, and guide plant operators toward initial investigations.

Apply PCA based MSP to monitor a fluid catalytic cracker unit using minute-level data from around 46 variables to detect faults and pinpoint furnace-related issues.

Explore partial least squares regression, its link to PCA, and how PLS uses score matrices T and U to connect X and Y and enable fault detection metrics.

Explore the mathematics behind PLS regression, including weight vectors, loading vectors, rotation vectors, deflation, and score construction, and learn how SP and t-squared monitor industrial processes.

Fit a PLS regression model on an LDPE reactor dataset using sklearn, explore variance captured by components, and assess predictive performance with about 90% r-squared.

Apply PLS-based fault detection to monitor industrial processes by computing t-squared, SP, and SPY metrics, establishing control limits, and flagging out-of-control events using training and test data.

Learn fault diagnosis in industrial processes using partial least squares (pls) and principal component analysis, perform contribution analysis on fault detection metrics, and identify variables via contribution plots for troubleshooting.

Apply PLS to monitor a refinery furnace, detecting disturbances and isolating faults with multivariate data from three manipulated and five output variables under MPC control.

Take your process monitoring model to end users with a web-based tool. Build a Flask app that serves PCA-based health assessments, Q and T-squared metrics, and HTML templates for dashboards.

Explore extensions of classical PCA and PLS for dynamic, nonlinear, and multi-cluster industrial process monitoring, including dynamic PCA, kernel PCA, kernel PLS, and cluster-based models.

Recap univariate and multivariate monitoring techniques—Shewhart and Cusum charts, PCA, and PLS regression—covering fault detection, fault diagnosis, and tool deployment for plant operators.