Reservoir Computing for Detection of Steady State in Performance Tests of Compressors

TL;DR

The paper tackles reducing the duration of industrial compressor performance tests by detecting steady-state entrance using Reservoir Computing with a novel self-organized subspace projection. By clustering the initial episode behavior with $k$-means and feeding a binary cluster-driven input to the reservoir, the approach constrains reservoir trajectories and achieves robust steady-state detection across diverse reservoir parameters. Evaluated on a large, heterogeneous dataset from thousands of compressor tests, the method substantially saves time per test while maintaining high predictive accuracy (AUC up to $\approx 0.99$) and low false positives. This work demonstrates practical impact for plant productivity and offers a data-driven, model-free alternative to handcrafted features or model-specific normalizations; it also points toward broader applicability of RC with subspace projection in time-series detection tasks.

Abstract

Fabrication of devices in industrial plants often includes undergoing quality assurance tests or tests that seek to determine some attributes or capacities of the device. For instance, in testing refrigeration compressors, we want to find the true refrigeration capacity of the compressor being tested. Such test (also called an episode) may take up to four hours, being an actual hindrance to applying it to the total number of compressors produced. This work seeks to reduce the time spent on such industrial trials by employing Recurrent Neural Networks (RNNs) as dynamical models for detecting when a test is entering the so-called steady-state region. Specifically, we use Reservoir Computing (RC) networks which simplify the learning of RNNs by speeding up training time and showing convergence to a global optimum. Also, this work proposes a self-organized subspace projection method for RC networks which uses information from the beginning of the episode to define a cluster to which the episode belongs to. This assigned cluster defines a particular binary input that shifts the operating point of the reservoir to a subspace of trajectories for the duration of the episode. This new method is shown to turn the RC model robust in performance with respect to varying combination of reservoir parameters, such as spectral radius and leak rate, when compared to a standard RC network.

Figures (12)

• Figure 1: Simplified schematic of the test rig.
• Figure 2: Reservoir Computing (RC) network. The reservoir is a non-linear dynamical system usually composed of recurrent sigmoid units. Solid lines represent fixed, randomly generated connections, while dashed lines represent trainable or adaptive weights.
• Figure 3: RC network with inputs divided in original inputs $u_1$ and binary inputs $u_2$.
• Figure 4: General overview of the diversity of dynamical behaviors and range for the normalized cooling capacity during an initial time period corresponding to the first 80 samples (i.e., $n_I=13$ minutes). Only 26 performance tests (from different compressor models) from the training dataset are shown.
• Figure 5: Generating a priori information $u_2$ for reservoir subspace projection through unsupervised learning in the very beginning of the episode. The vertical axis in the top plot represents any input signal $u_1$ (in our case, the cooling capacity). $u_c$ is the $(n_I \times 1)$-dimensional input to the clustering model.