Fault detection and diagnosis in air handling using data-driven methods

Highlights

• A model similar to RP-1312 model was considered for AHUs Using HVACSIM + software.

• In addition to RBF, PCA and KPCA methods were used for fault diagnosis.

• Diagnosing nine sensor and actuator faults simultaneously.

• Better results than previous works.

• The number of selected variables is optimal, so the computational time of the algorithm has reduced significantly.

Abstract

Actuator and sensor faults are inevitable in air handling units. It may cause loss of energy and reduction in fresh air quality and may endanger human life in some cases like, e.g., the operation room. This paper presents a fault detection and diagnosis method in air handling units. At first, a model similar to the RP-1312 model was considered for air handling units. Some fault signals were then applied to the system, and finally, data were obtained to be used in fault detection block. Due to the high dimensions and volume of the data, the problem dimensions should be reduced while maintaining the quality of detection. One of the contributions of this paper is to examine the effect of various variables on fault diagnosis performance and, while retaining excellent variables, eliminates inappropriate and recessive ones. An optimal selection of the proper variables for fault detection makes the computational time of the algorithm significantly reduced. Considering abundant data and their nonlinear nature, support vector machines technique and radial basis function neural network methods were used for fault detection and diagnosis, respectively. In addition to using the radial basis function neural network method, the principal component analysis technique and kernel principal components analysis, which is the nonlinear generalization of the principal components, were used for fault diagnosis. The other contribution of this paper is to detect the sensor and actuator faults simultaneously. The simulation results show that the proposed method accurately detects and diagnoses faults and has better results than previous works.

Introduction

In 1971, research into fault-tolerant control systems began and led to the design of fault tolerant-control for flight systems in 1985 [1]. In recent years, the demand for these types of systems has led to the emergence of a variety of fault detection and diagnosis systems [2].

Over recent decades, the study of faults in heating, ventilation, and air conditioning (HVAC) systems has received lots of attention because energy consumption is increased due to faulty HVAC systems.

Air Handling Units (AHUs) are one of the most extensive HVAC equipment in large commercial buildings. Fault detection and diagnosis methods are used to detect and diagnose abnormal conditions, faults, and system function impairment before further damage. It is estimated that between 5% and 30% of any building's annual energy consumption is unknowingly wasted due to equipment malfunctioning, such as air AHU [3]. Faults may occur incrementally over time or suddenly in actuators, system, or sensors. It should be noted that each fault has its own specific symptoms.

In order to design fault detection and diagnosis system in AHU, model-based [4] and data-driven methods [5,6] are used. Among data-driven methods are various techniques such as Principal Component Analysis (PCA) [[7], [8], [9]], Support Vector Machines (SVMs) [10,11], Active Function Testing (AFT) [12] It is also possible to use methods such as Kalman filter [13], Exponentially Weighted Moving Average (EWMA) [14], to detect and diagnose a fault.

Wang and Fu used two PCA models to establish thermal equilibrium and pressure flow in AHU. Sedimentation in coils, stuck dampers, failure of the fans, stuck fan blade, and the slip of the supply air fan belt were considered as the fault. These faults affect the PCA output; in fact, they do not eliminate thermal equilibrium but increase energy consumption [15].

Tashtush et al. presented a dynamic model of HVAC for a region that includes heating and cooling coil, dehumidifier, humidifier, duct, fan, and mixing box. The purpose of the controller design is to reduce energy consumption and improve indoor environmental conditions. A system capable of effectively controlling the disturbances in the shortest possible time with the least error was therefore designed. Eventually, it led to a PID design with the least possible fluctuations [16].

The effects of faults lead to improper function of AHU and increased energy consumption [17]. Y.Yu et al. reviewed and summarized faults and their detection, diagnosis, and isolation methods. They considered ten desirable features for fault detection, diagnosis, and isolation methods. These features include the possibility of quick fault detection and diagnosis, isolability, robustness, identifiability, classification error estimate, adaptability, explanation facility, modeling requirements, storage and computations, and the ability to diagnose several faults. It should be noted that no method has all features together, and fault detection method is selected based on needs [18].

Currently, one of the most widely used and effective AHU models is the RP-1312 AHU model that has been upgraded to ASHRAE projects (RP-825 and RP-1194 models) [19]. ASHRAE RP-1312 was implemented in the test center at the energy resource station (ERS). The RP-1312 model conducted several on-site tests to simulate the dynamic behaviors of a single-duct dual-fan VAV-AHU system serving four-building zones under various seasonal conditions [20]. Using HVACSIM + software for simulation, Lee and Van combined two new components to this model and reached some outcomes by experiments that validated the new model [21,22].

In the presented paper, data-driven detection and diagnosis methods such as SVMs, PCA, kernel principal components analysis (KPCA), and radial basis function (RBF) neural network were used on a model similar to RP-1312 model for fault detection, and fault diagnosis.

The structure of the work is the following, in Section 2, the modeling of the fault in AHU is presented, while the proposed fault detection and fault diagnosis methods are described in Section 3 Fault detection methods, 4 Fault diagnosis methods, respectively. Following that, Section 5 shows the simulation results, and finally, the calculations are described in Section 6.