We can help train your staff, audit existing AI models for accuracy and robustness, evaluate your AI processes, and help you deploy AI within your power generation organization.
When applied properly, AI has a multitude of powerful uses. It can be used to make extremely complex physical system models faster and more portable. It can be used to build predictive models from data where no physical model exists. AI is also extremely useful at helping to understand complex relationships where measurements are often unreliable, or come with a high degree of uncertainty. Our team regularly applies different AI techniques to speed up calibration of gas turbine performance models, identify anomalous combustion dynamics to detect impending failures, and to generating fault signatures for monitoring and diagnostics. AI can be used to greatly increase productivity, especially for repeatable engineering tasks.
Common Types of Machine Learning
Artificial Neural Network
Artificial Neural Network (ANN) are designed to mimic the connection of neurons in the human brain. ANNs is that this can adapt to discrete and non-linear responses and handle both discrete and continuous inputs simultaneously. This method is computationally efficient and portable once trained.
Clustering algorithms are typically used in APRI M&D software. It works by identifying clusters of common data points in multidimensional space. This method is good for unsupervised learning. Clustering algorithms are especially useful when functional form of data is not known or hard to define. The ability to generalize comes at the cost of accuracy and detectability. Common types of clustering algorithms are K-means, Hierarchical, and Normal Mixtures.
Classification Algorithms (Advanced Pattern Recognition)
Classification algorithms predict class membership based on input data. This method is conceptually similar to clustering, except that groups are tagged in advanced. Common types of algorithms are including Logistic Regression, Naive Bayes Classifier, K-Nearest Neighbors, Decision Trees, and Neural Networks. All algorithms predict probability that certain set of inputs belongs to the specific class.
Bayesian Learning is a real-world example of Bayesian Networks. This method is used in model calibration, diagnostics, and model updating. Bayesian Learning is flexible and good for mixed data sets, and suitable for both discrete and continuous data.