Wavelet-based analysis for fouling diagnosis of an aircraft heat exchanger

Abstract

Method for estimating a fouling condition of a heat exchanger (102) of an aircraft, comprising: Received at the heat exchanger (102) from drafts from an aircraft engine and from stagnant air; Connecting sensors at different positions with the heat exchanger (102) to measure various parameters related to the heat exchanger (102), wherein the sensors include a pressure sensor configured to measure an inlet pressure of the bleed air into the heat exchanger, a temperature sensor configured to measure a temperature associated with the heat exchanger (102), and an angular velocity sensor configured to measure an angular velocity of a shaft of a cooling turbine (106); Receiving training sensor measurements of a parameter relating to the operation of the heat exchanger (102); Applying a wavelet transformation to the training sensor measurements to obtain wavelet data; Performing wavelet data reduction to obtain a representative feature that indicates the training sensor measurements; Training a classifier (314) to assign a suitable fouling class to the representative feature; and Applying the trained classifier (316) to test sensor measurements from the heat exchanger (102) to assign a fouling class to the test sensor measurements in order to estimate the fouling condition of the heat exchanger (102); Evaluating the performance of the classifier (316) to reduce the number of false alarms and missed detections, wherein the evaluation includes the use of a K-fold cross-validation algorithm; Outputting a diagnosis of the fouling condition of the heat exchanger (102) for display on a screen (210); Performing maintenance on the aircraft's heat exchanger (102) based on the condition of the heat exchanger (102).

Description

GENERAL STATE OF THE ART

[0001] The present invention relates to aircraft environmental control systems (ECS), and more particularly to a system and method for estimating the fouling condition of a heat exchanger of an ECS.

[0002] An aircraft's environmental control system (ECS) performs two basic functions: i) to provide a supply of fresh air to the cabin, and ii) to provide thermal control and air pressure within an aircraft cabin to ensure the comfort of crew and passengers. One component of the ECS is a heat exchanger, which maintains the above two functions by properly cooling engine bleed air. Aircraft heat exchangers can experience a degradation in performance due to a phenomenon known as fouling, which occurs when there is an undesirable buildup of foreign substances, dirt, and / or organisms contaminating the surface of the heat exchanger fins. Fouling degrades the performance of the heat exchanger by reducing its thermal efficiency, thereby lowering the heat flux and placing additional stress on the heat exchanger's cooling mechanism.Due to the frequency with which fouling occurs, heat exchangers generally require periodic maintenance, causing economic losses and unwanted disruption to aircraft operations. Therefore, there is a need to develop a real-time diagnostic tool for detecting and diagnosing heat exchanger fouling on-site.

[0003] Because fouling occurs in complex environments where parameters such as altitudes, ambient temperatures, and passenger load can vary, diagnosing fouling conditions during ground maintenance operations can be difficult.

[0004] US 7 290 450 B2 describes a diagnostic device for detecting contamination and corrosion in process equipment, comprising: at least one detector for measuring a frequency response, which is coupled to the process equipment, wherein the frequency response is related to contamination or corrosion of the process equipment; a transducer coupled to a process element and configured to resonate the process element; and a diagnostic circuit coupled to the at least one detector and the transducer, wherein the diagnostic circuit is configured to resonate the process element by means of a command sent to the transducer and to process the measured frequency response from the at least one detector.The diagnostic circuit is further configured to compare the measured frequency response with a stored reference frequency response and generate a diagnostic signal indicating the condition of the process equipment with respect to contamination or corrosion based on a deviation of the measured frequency response from the stored frequency response.

[0005] US 2013 / 0 013 145 A1 describes an electronic controller for a subsystem of an air management system, the controller comprising: an area conversion module, the area conversion module comprising: a setpoint input for the engine bleed air outlet pressure; a measurement input for the engine bleed air outlet pressure; a setpoint input for a subsystem parameter; a measurement input for a subsystem parameter; a measurement input for the subsystem outlet pressure; a setpoint sub-module for generating a calculated area setpoint output as a function of at least the setpoint input for the engine bleed air outlet pressure, the setpoint input for the subsystem parameter and the measurement input for the subsystem outlet pressure;and a measurement sub-module for generating a calculated area measurement output as a function of at least the engine bleed air outlet pressure measurement input, the subsystem parameter measurement input, and the subsystem outlet pressure measurement input; and a subsystem area control module for generating a control output for the subsystem, the control output being a function of the area setpoint output calculated by the area conversion module and the area measurement output calculated by the area conversion module.

[0006] WO 2015 / 082 913 A2 describes a fuel system for an aircraft, comprising: a fuel storage system having a fuel tank; a heat exchanger; a fuel cooling circuit configured to conduct fuel from the storage system through the heat exchanger and back to the storage system; and a pump configured to pump fuel through the cooling circuit.

[0007] US 2005 / 0 177 317 A1 describes a computer system comprising: a process manager; an input module connected to the process manager and providing first data from a first sensor device and second data from a second sensor device to a computing device, the first and second sensor devices being connected to the input module via a computer network; and a pattern recognition module connected to the process manager and processing the first and second data using a pattern recognition algorithm to classify or identify a substance.

[0008] CN 1 996 190 A describes an industrial production process diagnostic system based on small-signal analysis. It comprises an industrial process object connected to a locally installed smart meter, a DCS system, and its higher-level control unit. The DCS system consists of a data interface, a control station, and a database. The smart meter, the DCS system, and the higher-level control unit are connected in sequence. The higher-level control unit consists of a standardized processing module, a small-signal resolution module, a pivot point analysis function module, a small-signal reconstruction module, a classifier module based on support vector machines, and a diagnostic evaluation module. BRIEF DESCRIPTION OF THE INVENTION

[0009] The present invention is specified in independent claims 1 and 8. Particular embodiments emerge from the respective dependent claims. In one aspect, a method for estimating a fouling condition of a heat exchanger of an aircraft includes: obtaining training sensor measurements of a parameter related to the operation of the heat exchanger; applying a wavelet transform to the training sensor measurements to obtain wavelet data; performing data reduction on the wavelet data to obtain representative features indicative of the training sensor measurements; training a classifier to assign appropriate fouling classes to the representative features; and applying the trained classifier to test sensor measurements from the heat exchanger to assign a fouling class to the test sensor measurements to estimate the fouling condition of the heat exchanger.

[0010] In another aspect, an apparatus for estimating a fouling condition of a heat exchanger of an aircraft includes: a model of the heat exchanger; one or more sensors configured to obtain training sensor measurements of a parameter related to operation of the model of the heat exchanger; and a processor configured to: wavelet transform the training sensor measurements to obtain wavelet data; perform data reduction on the wavelet data to obtain representative features indicative of the training sensor measurements; train a classifier to assign appropriate fouling classes to the representative features; and apply the trained classifier to test sensor measurements to assign a fouling class to the test sensor measurements to estimate the fouling condition of the heat exchanger. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The subject matter regarded as the invention is expressly pointed out and expressly claimed in the claims at the conclusion of the specification. The foregoing and other features and advantages of the present invention will become apparent from the following detailed description taken in combination with the accompanying drawings, in which: Fig. 1 shows an exemplary aircraft environmental control system (ECS) that may be diagnosed using the methods described herein; Fig. 2 a system suitable for determining the fouling condition of the ECS using the data from the sensors of Fig. 1 to estimate the measurements obtained; Fig. 3A and Fig. 3B is a schematic diagram of an illustrative method for estimating a fouling condition of the ECS in one embodiment of the present invention; and Fig. 4 different wavelet representations in a wavelet domain of sensor data for a heat exchanger model. DETAILED DESCRIPTION OF THE INVENTION

[0012] The present invention estimates a fouling level of an aircraft's environmental control system (ECS) using wavelet analysis and a method for data reduction of simulated or training data from the ECS. The method disclosed herein utilizes a classifier trained using features obtained by wavelet analysis of the time-domain measurements and data reduction of this wavelet-transformed data. The classifier can then learn behavior of a model of a heat exchanger of the ECS under various fouling conditions, including normal conditions where no fouling is present. The trained classifier can then be applied to real-time data from the ECS to determine the fouling level of the ECS. Accordingly, the methods disclosed herein can be used to operate the ECS, as well as to monitor and / or regulate the operation of the ECS.

[0013] Fig. Figure 1 shows an exemplary aircraft environmental control system (ECS) 100 that can be diagnosed using the methods disclosed herein. The ECS 100 supplies pressurized and temperature-controlled air to a cabin of an aircraft. The ECS 100 includes a heat exchanger 102, a condenser (CND) 104, and an air cooling turbine (ACM) 106.

[0014] Heat exchanger 102 receives bleed air from an aircraft engine as well as ram air. In one embodiment, heat exchanger 102 is a fin-and-plate heat exchanger constructed as a series of parallel aluminum plates stacked one upon another. The spaces between these plates consist of chambers with square fins that serve to transfer heat between the bleed air (hot) and the ram air (cold). Heated ram air is then circulated through a first fluid line 120 between heat exchanger 102 and ACM 106 to drive ACM 106. Spent air from ACM 106 is returned to heat exchanger 102 through a second fluid line 122.

[0015] The ACM 106 is an internal cooling unit for the aircraft and uses ram air as a medium to cool conditioned bleed air, which is ultimately used to cool the cabin. The ACM 106 includes a compressor 108, a first turbine 110, and a second turbine 112. The compressor 108 is used to compress the ram air to increase its temperature and pressure. The first turbine 110 and the second turbine 112 use the thermal energy of the compressed air to rotate a shaft on which the compressor 108 and the first and second turbines 110 and 112 are located.

[0016] Condenser 104 enables heat transfer by acting as a catalyst for converting steam into a liquid. First turbine 110 can provide gas or steam to condenser 104 for condensation. The condensed heat can be stored in a collection unit 114, from which it is passed to a reheater 116, which is in thermal communication with condenser 104. The reheated steam can then be used at second turbine 112.

[0017] The heat exchanger 102, the compressor 108, and the two turbines 110 and 112 are subject to various degradation processes. The heat exchanger 102 is subject to fouling due to surface contamination. Fouling occurs due to the growth of living organisms on the surface of the heat exchanger 102 or due to the deposition of inanimate substances, which may be organic or inorganic. The effects of fouling include the contamination of the heat exchanger surfaces, which reduces the flow of ram air through the plate fins, thereby reducing the amount of heat transfer. This degradation impairs the thermal efficiency of the heat exchanger 102 and the heat flow across the heat exchanger 102.

[0018] Sensors are connected to the ECS 100 at various positions to measure various parameters related to the ECS 100 and / or the heat exchanger 102. The pressure sensor P inmeasures an inlet pressure of the bleed air into the heat exchanger 102. The temperature sensor TH X1 measures a primary heat exchanger outlet temperature, ie the temperature of gas circulating from heat exchanger 102 to compressor 108. The temperature sensor TH X2 measures a second heat exchanger temperature, ie the temperature of spent gas circulating from the compressor 108 to the heat exchanger 102. The temperature sensor T CND measures a condenser inlet temperature. The temperature sensor T aus measures a temperature at an outlet of the ACM 106. The angular velocity sensor ω ACM measures an angular velocity of the ACM 106, ie the shaft of the ACM 106.

[0019] Fig. Figure 2 shows a system 200 suitable for determining the fouling condition of the ECS 100 using the information provided by the sensors of Fig. 1. The system 200 includes a control unit 202 that includes a processor 204 and a storage device 206. The storage device 206 may include a non-transitory computer-readable medium, such as a read-only memory device. Programs or instructions 208 may be stored in the storage device 206 that, when read by the processor 204, enable the processor 204 to estimate the fouling condition of the ECS 100. The processor 204 receives the measurements obtained from the sensors Fig. 1 and outputs a diagnosis of a fouling condition of the ECS 100 for display on a screen 210. The diagnosis can be viewed, for example, by the pilot of the aircraft, allowing the pilot to plan a maintenance procedure or take appropriate action during the flight. The present invention's method for estimating the fouling condition is further described below.

[0020] Fig. 3A and Fig. 3B shows a schematic diagram 300 of an illustrative method for estimating a fouling condition of the heat exchanger 102 of the ECS 100 in one embodiment of the present invention. The method includes a training phase (see Fig. 3A) and a test phase (see Fig. 3B). The training phase uses a set of training sensor data, which may be simulated sensor data generated by a model of the heat exchanger 102 or data from the operation of a heat exchanger 102 in a laboratory or under controlled conditions. The training phase results in a trained classifier 316, which is then used in the testing phase with real-time sensor data obtained from a heat exchanger 102. The training phase and the testing phase may be performed using the processor 204 of Fig. 2 be carried out or executed.

[0021] The training phase includes a feature extraction process 302, a classifier training process 304, and a classifier evaluation process 306. The feature extraction process ("feature extractor") 302 receives training sensor data 312. The feature extraction process 302 includes a wavelet analysis process ("wavelet analyzer") 308 and a data reduction process ("data reducer") 310. In one embodiment, the data reduction process 310 uses principal component analysis (PCA). The wavelet analysis process 302 performs a wavelet transform on the training data 312 to obtain corresponding wavelet data in a wavelet domain. The wavelet data includes temporal and spectral information of the sensor measurements. The wavelet data from the wavelet analysis process 302 is input to the data reduction process 310.The wavelet data may be subjected to principal component analysis (in data reduction process 310) to reduce the dimensionality of the data, thereby enabling the extraction of useful features related to principal components of the data. Although principal component analysis (PCA) is disclosed herein in an exemplary embodiment, any form of data reduction may be used. Such data reduction methods generally transform the data to obtain one or more representative features of the wavelet data in a reduced dimensional space. The feature extraction process 302 sends the representative features to the classifier training process 304.

[0022] The classifier training process 304 uses the representative features to classify the fouling condition of the heat exchanger 102. In an illustrative embodiment, the fouling condition may belong to one of three classes. A first class (e.g., the nominal class) includes fouling levels from 0% to 25%, corresponding to minimal or no fouling. A second class (e.g., the warning class) includes fouling levels from 25% to 50%, corresponding to a degree of fouling that requires maintenance at the earliest possible opportunity. A third class (e.g., the critical class) includes fouling levels from 50% to 100%, corresponding to a degree of fouling that requires immediate maintenance. Although three fouling classes are described in the above description, any number of fouling classes may be defined in alternative embodiments.The fouling classes of the heat exchanger 102 can be known or defined in advance.

[0023] The classifier training process 304 generates a trained classifier 316 suitable for local use during the testing phase. The trained classifier 316 is trained using a model of the ECS 100 operating under various fouling conditions of the heat exchanger 102, such as a "nominal fouling" condition and various predefined fouling levels. The trained classifier 316 is trained incorporating uncertainties in the data. Such uncertainties may include process or measurement disturbances and unmodeled physical dynamics. To classify the trained classifier 316, the wavelet analysis process 308 first converts time-domain sensor data obtained from a model of the ECS into the wavelet domain. Noise reduction and enhancement of patterns generated under different fouling conditions can be performed in the wavelet domain.Subsequently, the representative features (e.g., principal components) are extracted from the wavelet domain using the data reduction process 310. These features can be classified into different fouling classes, which are then used by the classifier training process 304 to train the classifier 314 to obtain the trained classifier 316. In one embodiment, the classifier 314 uses a k-nearest neighbor (k-NN) algorithm as a decision function to classify the fouling severity of the heat exchanger given a set of training sensor measurements. The trained classifier 316 can then be implemented for real-time fouling diagnosis of the heat exchanger 102.

[0024] Fig. Figure 4 shows various graphical wavelet surface representations in a wavelet domain for a heat exchanger model. The wavelet surface representations are shown in ascending order of fouling severity. The trained classifier 316 uses the features of the wavelet surface representations obtained from the measurements in the heat exchanger model to determine the fouling extent associated with the measurements.

[0025] The classifier evaluation process 306 may be used to evaluate the performance of the classifier 314. In one embodiment, the performance of the classifier 314 may be evaluated using a K-fold cross-validation (CV) algorithm 318. The sensor data is randomly divided into groups for each iteration, with a portion of the groups used to train the classifier 314 and the remaining groups used to test the classifier 314. Multiple iterations of the K-fold CV algorithm 318 may be tabulated in a confusion matrix 320, which summarizes the performance of the classifier by correctly assigning one or more classes to the representative features of the training sensor measurements.The confusion matrix 320 can be used to determine the percentage of correct classification of all test points used by the iterations of the K-fold CV algorithm 318.

[0026] Referring to Fig. 3B, the trained classifier 316 is used in the test phase to determine a fouling level of a heat exchanger 102. Sensor data 336 (also referred to as “test sensor measurements”), such as the temperature, pressure, and angular velocity data of the sensors from Fig. 2, are received from the heat exchanger 102 and sent to a local feature extraction process 330. The local feature extraction process 330 includes a wavelet analysis process 332 and a data reduction process 334. The wavelet analysis process 332 and the data reduction process 334 essentially perform the same processes as those performed by the feature extraction process 302. Fig.3A. The result of these processes are the representative features (i.e., principal components) that are sent to the trained classifier 316, which makes a class decision based on the sensor data 336 to indicate a fouling class of the heat exchanger 102. The class decision, in various embodiments, may include determining a fouling level or recommending scheduling maintenance for the heat exchanger / environmental control system or sending a notification to a pilot.

[0027] The present invention therefore provides early detection of heat exchanger fouling and identification of its severity. Fouling diagnosis can be performed in the presence of various uncertainties in the system to minimize false alarms and missed detections. Maintenance of the aircraft heat exchanger can thus be performed based on the condition of the heat exchanger rather than using periodic maintenance procedures. Condition-based maintenance leads to increased utilization of aircraft heat exchangers in operating condition and thus additional aircraft utilization.

[0028] Although the invention has been described in detail in connection with a limited number of embodiments, it is to be understood that the invention is not limited to such disclosed embodiments. Rather, the invention may be modified to include any number of arrangements of variations, changes, substitutions, or equivalents not described hereinabove, but which are equally consistent with the spirit and scope of the invention. Furthermore, while various embodiments of the invention have been described, it is to be understood that aspects of the invention may include only a few of the described embodiments. Accordingly, the invention is not to be considered limited by the foregoing description, but is limited only by the scope of the appended claims.

Claims

[1] A method for estimating a fouling condition of a heat exchanger (102) of an aircraft, comprising: Receiving, at the heat exchanger (102), draft air from an aircraft engine and ram air; Connecting sensors at different positions to the heat exchanger (102) to measure various parameters related to the heat exchanger (102), the sensors comprising a pressure sensor configured to measure an inlet pressure of the bleed air into the heat exchanger, a temperature sensor configured to measure a temperature associated with the heat exchanger (102), and an angular velocity sensor configured to measure an angular velocity of a shaft of a cooling turbine (106); Obtaining training sensor measurements of a parameter relating to the operation of the heat exchanger (102); Applying a wavelet transform to the training sensor measurements to obtain wavelet data; Performing data reduction on the wavelet data to obtain a representative feature indicative of the training sensor measurements; Training a classifier (314) to assign a suitable fouling class to the representative feature; and Applying the trained classifier (316) to test sensor measurements from the heat exchanger (102) to assign a fouling class to the test sensor measurements to estimate the fouling condition of the heat exchanger (102); Evaluating a performance of the classifier (316) to reduce a number of false alarms and missed detections, the evaluation comprising using a K-fold cross-validation algorithm; Outputting a diagnosis of the fouling condition of the heat exchanger (102) for display on a screen (210); Performing maintenance of the aircraft's heat exchanger (102) based on the condition of the heat exchanger (102). [2] The method of claim 1, wherein the heat exchanger (102) is included in an environmental control system having a cooling turbine, and the parameter further comprises at least one of: (i) heat exchanger outlet temperature; (ii) condenser inlet temperature; (iii) environmental control system outlet temperature; (iv) environmental control system inlet pressure; and (v) an angular velocity of a cooling turbine. [3] Method according to claim 1 or 2, wherein the fouling class of the heat exchanger (102) is known in advance. [4] A method according to any one of the preceding claims, further comprising classifying the representative feature using a pattern classification algorithm. [5] The method of claim 4, wherein training the classifier (314) further comprises applying a cross-validation algorithm to test the ability of the classifier (314) to properly classify the representative feature. [6] The method of any preceding claim, further comprising obtaining the training sensor measurements from a model of the heat exchanger (102) under various fouling conditions. [7] The method of claim 6, wherein training the classifier (314) further comprises setting fouling classification categories. [8] Apparatus for estimating a fouling condition of a heat exchanger (102) of an aircraft, comprising: a model of the heat exchanger (102); one or more sensors configured to obtain training sensor measurements of a parameter related to operation of the model of the heat exchanger (102), the sensors comprising a pressure sensor configured to measure an inlet pressure of bleed air into the heat exchanger, a temperature sensor configured to measure a temperature associated with the heat exchanger (102), and an angular velocity sensor configured to measure an angular velocity of a shaft of a cooling turbine (106); and a processor configured to: apply a wavelet transform to the training sensor measurements to obtain wavelet data; Perform data reduction of the wavelet data to obtain a representative feature, to train a classifier (314) to assign a suitable fouling class to the representative feature, and applying the trained classifier (316) to test sensor measurements from the heat exchanger (102) to assign a fouling class to the test sensor measurements to estimate the fouling condition of the heat exchanger (102); evaluate a performance of the classifier (316) to reduce a number of false alarms and missed detections, the evaluation comprising using a K-fold cross-validation algorithm; to output a diagnosis of the fouling condition of the heat exchanger (102) for display on a screen (210). [9] The apparatus of claim 8, wherein the heat exchanger (102) is included in an environmental control system having a cooling turbine, and the parameter further comprises at least one of: (i) heat exchanger outlet temperature; (ii) condenser inlet temperature; (iii) environmental control system outlet temperature; (iv) environmental control system inlet pressure; and (v) an angular velocity of a cooling turbine. [10] The apparatus of claim 8 or 9, wherein the processor is further configured to estimate the fouling condition by subjecting a representative feature to a pattern classification algorithm to classify the representative feature. [11] Device according to one of claims 8 to 10, wherein the fouling class of the heat exchanger (102) is known in advance. [12] The apparatus of claim 11, wherein training the classifier (314) further comprises applying a cross-validation algorithm to test the ability of the classifier (314) to properly classify the representative feature. [13] The apparatus of any one of claims 8 to 12, wherein the processor is further configured to train the classifier (314) using the model of the heat exchanger (102) under different fouling conditions. [14] The apparatus of any one of claims 8 to 13, wherein training the classifier (314) further comprises setting fouling classification categories.

Images