KR20170031985A - Fault detection and diagnostics method of air-conditioning system - Google Patents
Fault detection and diagnostics method of air-conditioning system Download PDFInfo
- Publication number
- KR20170031985A KR20170031985A KR1020150129632A KR20150129632A KR20170031985A KR 20170031985 A KR20170031985 A KR 20170031985A KR 1020150129632 A KR1020150129632 A KR 1020150129632A KR 20150129632 A KR20150129632 A KR 20150129632A KR 20170031985 A KR20170031985 A KR 20170031985A
- Authority
- KR
- South Korea
- Prior art keywords
- data
- machine learning
- virtual
- abnormal
- fault
- Prior art date
Links
Images
Classifications
-
- F24F11/0086—
-
- F24F11/0009—
-
- F24F11/001—
-
- F24F11/006—
-
- F24F11/02—
-
- F24F2011/0063—
-
- F24F2011/0091—
Abstract
The present invention has the advantage of being able to more accurately and objectively determine and diagnose faults through machine learning methods using virtual data and actual data of an air conditioning system.
Description
The present invention relates to a fault detection and diagnosis method of an air conditioning system, and more particularly, to a fault detection and diagnosis method of an air conditioning system for detecting and diagnosing faults using a machine learning method.
Generally, the air conditioning system of a building is also referred to as an HVAC system (heating, ventilation and air conditioning system), and the facilities are becoming larger and more complicated due to the heightened and intelligent buildings. As the installation of the air conditioning system becomes larger and the control becomes complicated, the economic loss and the risk can be increased when the failure occurs. In the case of a gradual failure of the air conditioning system, the performance of the system is deteriorated over a long period of time, which may result in waste of energy and deterioration of the system stability. Accordingly, there is an increasing interest in performing fault detection and diagnosis of the air conditioning system in recent years.
However, when a method of regularizing the characteristic values set by the manager and detecting a fault signal when the rule is not used is used, it is not reflected when the characteristic is changed due to the deterioration of the facility or the like, And the information about the actual data value is not reflected, resulting in a problem that the accuracy is lowered.
An object of the present invention is to provide a fault detection and diagnosis method of an air conditioning system capable of more accurately detecting and diagnosing faults.
A fault detection and diagnosis method of an air conditioning system according to the present invention includes the steps of generating virtual normal data and virtual abnormal data for a plurality of predetermined faults through simulation of an air conditioning system; Generating a machine learning diagnostic model that determines whether the virtual normal data and the virtual abnormal data are normal or abnormal with respect to the failures by mechanically learning the virtual normal data and the virtual abnormal data using a predetermined machine learning algorithm; Wherein the plurality of sensors measure data in real time and generate real time data when the air conditioning system is operated; And applying the real-time data to the machine learning diagnostic model to derive a diagnostic result from the machine learning diagnostic model to diagnose whether the fault is a steady state in which the fault has not occurred or an abnormal state in which the fault has occurred.
According to another aspect of the present invention, there is provided a method for detecting and diagnosing a fault in an air conditioning system, comprising: simulating an air conditioning system to create a virtual state in which a predetermined fault has not occurred and an abnormal state in which the fault has occurred, Generating virtual data by filtering the virtual data; Generating a machine learning diagnostic model that determines whether the virtual normal data and the virtual abnormal data are normal or abnormal with respect to the failures by mechanically learning the virtual normal data and the virtual abnormal data using a predetermined machine learning algorithm; A plurality of sensors measuring data in real time and filtering measured data to generate real-time data in operation of the air conditioning system; Real-time data is applied to the machine learning diagnostic model to output the probability of occurrence of the failure from the machine learning diagnostic model, and if the output probability exceeds a predetermined reference value, diagnosis is made to diagnose that the failure has occurred Deriving a result; And the machine learning algorithm modifying the machine learning diagnostic model in accordance with the real-time data and the diagnostic result.
The present invention has the advantage of being able to more accurately and objectively determine and diagnose faults through machine learning methods using virtual data and actual data of an air conditioning system.
1 is a block diagram illustrating a process for generating a machine learning diagnostic model in a method for detecting and diagnosing faults in an air conditioning system in accordance with an embodiment of the present invention.
2 is a block diagram illustrating a process for diagnosing faults by applying real-time data to a machine learning diagnostic model.
3 is a flowchart illustrating a method for detecting and diagnosing a failure of an air conditioning system according to an embodiment of the present invention.
Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.
1 is a block diagram illustrating a process for generating a machine learning diagnostic model in a method for detecting and diagnosing faults in an air conditioning system in accordance with an embodiment of the present invention. 2 is a block diagram illustrating a process for detecting a failure using the machine learning diagnostic model generated in FIG. 3 is a flowchart illustrating a method for detecting and diagnosing a failure of an air conditioning system according to an embodiment of the present invention.
1 to 3, a fault detection and diagnosis method of an air conditioning system according to an embodiment of the present invention will be described as follows.
First, a diagnosis engine for detecting and diagnosing a failure of the air conditioning system simulates an air conditioning system. (S1) The diagnosis engine may be a control unit for controlling the air conditioning system, or a separate computer . The simulation may use a simulation model set similar to the configuration of the air conditioning system, and it is also possible to actually experiment and operate the air conditioning system. In this embodiment, a simulation model called a Dymola model will be used as an example.
Virtual normal data and virtual abnormal data for a plurality of faults are generated through the simulation (S2) (S3)
The step of generating virtual data including the normal data and the abnormal data includes a step S2 of artificially creating the virtual data and a step S3 of filtering the virtual data.
In the process of artificially creating the virtual data (DATA_A to N), a plurality of faults are set in advance, normal data for a steady state in which a predetermined plurality of faults are not generated, and abnormal data for a plurality of faults Lt; / RTI > The virtual data are learning data for performing machine learning described later. The virtual data is data obtained from a plurality of sensors included in the simulation model. These failures include anomalies of the respective equipments constituting the air conditioning system. As an example, the failures include reduced efficiency of the supply fan, leakage of the cooling coil valve, reduced efficiency of the heat exchanger, and simultaneous failure of the supply fan and the cooling coil valve. The virtual normal data is data obtained from a plurality of sensors included in the simulation model in a virtual steady state in which the failure has not occurred. The virtual abnormal data is data obtained from the sensors in a virtual abnormal state in which the failure occurs. The measured values obtained from the sensors are shown in Table 1. In this embodiment, a total of N failures are set in advance and an example will be described by generating virtual normal data and abnormal data for N failures, respectively. That is, the failure includes faults A through N, and the virtual data DATA_A through N include normal data and abnormal data for the fault A, normal data and abnormal data for the fault B, and the like. The normal data for the fault A is data in a state in which no fault A occurs, and the abnormal data for the fault A is data in a state in which a fault A occurs.
When the virtual data is generated, the generated virtual data is filtered. (S3) The process of filtering the virtual data is also referred to as data preprocessing. The virtual data may have irregular results due to the addition of noise due to the influence of the building or surrounding environment. Therefore, in order to minimize such irregularity, a normalization process is performed in which standardization of each data is standardized.
In addition, the virtual data is subjected to a binarization process to convert it into binary data in order to use a DBN (Deep Belief Network) algorithm described later.
In addition, Principal Component Analysis (PCA) algorithm is applied to reduce the dimension of the virtual data.
After filtering the virtual data as described above, machine learning is performed using a predetermined machine learning algorithm. (S4) The machine learning algorithm uses a DBN (Deep Belief Network) algorithm. The DBN algorithm can be used even when there is no information on the steady state or abnormal state in the data information obtained from the sensors. At this time, a plurality of different DBN algorithms (DBN_A to N) are applied to each of the plurality of faults (A to N).
(S5), the machine learning diagnostic models (Model_A to N) are generated for the plurality of faults (A to N), respectively, by using the DBN algorithm . For example, among a plurality of machine learning diagnostic models, Model A (Model A) refers to a model for diagnosing whether a fault A is in a steady state or an abnormal state. The machine learning diagnostic models (Model_A to N) can be modified in accordance with real-time data and diagnosis results to be described later.
As described above, when a plurality of machine learning diagnostic models (Model_A to N) are generated for a plurality of faults (A to N), the air conditioning system can be actually operated to apply the machine learning diagnostic model. S6)
When the air conditioning system is actually operated, the sensors provided in the air conditioning system measure data in real time. (S7) Measurement data measured by the sensors are shown in Table 1.
The measurement data measured by the sensors are subjected to a filtering process to generate real-time data. (S8) The filtering process is the same as the filtering process of the learning data. That is, the measurement data is normalized and the dimension is reduced through a principal component analysis algorithm.
When the real-time data is generated, the real-time data is input to the plurality of machine learning diagnostic models (Model_A ~ N). Wherein the air conditioning system is configured to detect whether the air conditioning system is in a normal state in which no failure occurs or in an abnormal state in which a failure has occurred from the machine learning diagnosis model by inputting the real time data to the plurality of machine learning diagnosis models (Model_A ~ (S9). That is, the plurality of machine learning diagnostic models (Model_A to Model N) can calculate the probability of occurrence of the plurality of faults (A to N) And if the output probability exceeds a predetermined reference value, it is judged that a failure has occurred and is in an abnormal state. When there are a plurality of faults determined to be abnormal among the plurality of faults (A to N), the fault having the highest probability is preferentially diagnosed and derived from the diagnosis result. For example, if a probability that a first fault (A) occurs in the first model (Model_A) among the plurality of machine learning diagnostic models (Model_A to N) is output and the output probability exceeds a predetermined first reference value , It is detected that the first fault (A) is generated and is in an abnormal state. Also, if a probability that a second fault (B) is generated in the second model (Model_B) out of the plurality of machine learning diagnostic models (Model_A to N) is output and the output probability exceeds a predetermined second reference value, The second fault B is detected and it is detected that it is in an abnormal state. If both of the first fault A and the second fault B are generated and are all detected as abnormal, the diagnosis engine determines whether the first fault A occurs and the second fault B The probability of occurrence is compared with each other to diagnose a fault with a higher probability. If the probability that the first fault A is generated is higher than the probability that the second fault B is generated, it is determined that the first fault A has occurred, . Therefore, among the plurality of faults A to N, facilities associated with the fault having the highest probability can be checked first.
Accordingly, there is an advantage in that the fault can be detected and diagnosed through the mechanical learning method using the actual data of the air conditioning system, thereby making it possible to more accurately and objectively judge the abnormality as compared with the case where the administrator defines the set value.
In addition, the diagnostic engine may machine-learn the real-time data and the diagnostic result using the machine learning algorithm to modify the machine learning diagnostic model. That is, it is possible to use the real-time data and the diagnosis result as learning data.
The method may further include a Post processing step of post-processing the diagnosis result from the machine learning diagnostic model.
The post-processing step may process the results from the machine learning diagnostic model using a reservation method and a sliding window method.
The reservation method outputs a probability of occurrence of the plurality of faults (A to N), and reserves diagnosis of faults that are less than a predetermined reference value out of the respective outputs of the plurality of faults (A to N). Therefore, it is possible to suspend judgment on uncertain information about data that is difficult to diagnose.
The sliding window method derives a diagnostic result from the machine learning diagnostic model by extending the range for performing the machine learning in the machine learning diagnostic model by a predetermined range. That is, in making the diagnosis in the machine learning diagnosis model, not only the current data but also past data as much as the window size is considered. The larger the window size, the better the diagnostic performance, but the time delay can occur.
While the present invention has been described with reference to exemplary embodiments, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, but, on the contrary, is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims. Accordingly, the true scope of the present invention should be determined by the technical idea of the appended claims.
Claims (14)
Generating a machine learning diagnostic model that determines whether the virtual normal data and the virtual abnormal data are normal or abnormal with respect to the failures by mechanically learning the virtual normal data and the virtual abnormal data using a predetermined machine learning algorithm;
Wherein the plurality of sensors measure data in real time and generate real time data when the air conditioning system is operated;
Applying the real-time data to the machine learning diagnostic model to derive from the machine learning diagnostic model a diagnosis result that diagnoses whether the fault is a normal state in which the fault has not occurred or an abnormal state in which the fault has occurred, A method for detecting and diagnosing faults in a system.
Wherein the generating of the virtual normal data and the virtual abnormal data comprises:
Artificial virtual data for a normal state in which the failure has not occurred and an abnormal state in which the failure has occurred through the simulation;
And filtering the virtual data. ≪ RTI ID = 0.0 > 11. < / RTI >
Wherein the filtering the virtual data comprises:
And normalizing the data, and performing PCA (Principal Component Analysis) on the data.
Wherein the machine learning algorithm is a DBN (Deep Belief Network).
Wherein the generating of the real-
Measuring the data in real time by the sensors,
And filtering the measured data. A method of detecting and diagnosing faults in an air conditioning system, comprising:
Wherein the step of filtering the measured data comprises:
And normalizing the data, and performing PCA (Principal Component Analysis) on the data.
Further comprising: mechanically learning the real-time data and the diagnostic results therefrom to modify the machine learning diagnostic model.
The step of deriving the diagnostic result comprises:
And outputting probabilities of occurrence of each of the plurality of failures from the machine learning diagnostic model.
Comparing the probabilities of the plurality of failures with preset reference values, and determining that the failure is an abnormal state when the probability exceeds the reference value.
Comparing the probabilities of the faults judged to be abnormal if the plurality of faults judged to be in the abnormal state are compared with each other to derive the fault having the highest probability as the diagnosis result.
Further comprising a post-processing step of post-processing the diagnostic result from the machine learning diagnostic model.
The post-
And outputting a probability of occurrence of the failure, and reserving diagnosis of the failure if the output probability is less than a preset reference value.
The post-
And deriving a diagnostic result from the machine learning diagnostic model by increasing the range for performing the machine learning in the machine learning diagnostic model by a predetermined range.
Generating a machine learning diagnostic model that determines whether the virtual normal data and the virtual abnormal data are normal or abnormal with respect to the failures by mechanically learning the virtual normal data and the virtual abnormal data using a predetermined machine learning algorithm;
A plurality of sensors measuring data in real time and filtering measured data to generate real-time data in operation of the air conditioning system;
Real-time data is applied to the machine learning diagnostic model to output the probability of occurrence of the failure from the machine learning diagnostic model, and if the output probability exceeds a predetermined reference value, diagnosis is made to diagnose that the failure has occurred Deriving a result;
And the machine learning algorithm modifying the machine learning diagnostic model according to the real-time data and the diagnostic result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| KR1020150129632A KR20170031985A (en) | 2015-09-14 | 2015-09-14 | Fault detection and diagnostics method of air-conditioning system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| KR1020150129632A KR20170031985A (en) | 2015-09-14 | 2015-09-14 | Fault detection and diagnostics method of air-conditioning system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| KR20170031985A true KR20170031985A (en) | 2017-03-22 |
Family
ID=58497536
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| KR1020150129632A KR20170031985A (en) | 2015-09-14 | 2015-09-14 | Fault detection and diagnostics method of air-conditioning system |
Country Status (1)
| Country | Link |
|---|---|
| KR (1) | KR20170031985A (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108334999A (en) * | 2018-05-09 | 2018-07-27 | 山东交通学院 | The failure prediction method and system of fume hot-water type BrLi chiller |
| CN108647470A (en) * | 2018-05-29 | 2018-10-12 | 杭州电子科技大学 | A kind of localization method at the beginning of based on the leakage loss with depth belief network is clustered |
| KR20180130294A (en) * | 2017-05-29 | 2018-12-07 | 부경대학교 산학협력단 | Method for diagnosing machine fault based on sound |
| WO2019078515A1 (en) * | 2017-10-18 | 2019-04-25 | 삼성전자주식회사 | Data learning server, and method for generating and using learning model thereof |
| CN109882996A (en) * | 2019-01-25 | 2019-06-14 | 珠海格力电器股份有限公司 | A kind of method and apparatus of control |
| KR20190071503A (en) * | 2017-12-14 | 2019-06-24 | 한전케이디엔주식회사 | Enhanced learning based energy usage virtual data generation system |
| KR20190077358A (en) * | 2017-05-31 | 2019-07-03 | 신와 콘트롤즈 가부시키가이샤 | Status monitoring device, status monitoring method and program |
| KR20210061517A (en) * | 2019-11-19 | 2021-05-28 | 한국생산기술연구원 | Apparatus and method for fault diagnosis using fake data generated by machine learning |
| KR102268733B1 (en) * | 2020-06-02 | 2021-06-23 | 한국해양대학교 산학협력단 | Ship engine failure detection method and system |
| KR102281640B1 (en) * | 2021-03-24 | 2021-07-26 | 주식회사 유한테크 | AI Gas Leak Detection System with Self-Diagnosis Function and operating Method thereof |
| KR20210097412A (en) * | 2020-01-30 | 2021-08-09 | 전남대학교산학협력단 | Building Energy Failure Diagnosis and Analysis System with integrated Virtual sensor and Deep learning, and Building Energy Failure Diagnosis and Analysis Method |
| KR20220082597A (en) | 2020-12-10 | 2022-06-17 | 인천대학교 산학협력단 | Method and apparatus for sensing an energy system in a building |
| KR20230066874A (en) * | 2021-11-08 | 2023-05-16 | 중앙대학교 산학협력단 | Real-time self-adaptive steady state diagnosis method of heat pump system, recording medium and steady state diagnosis device for performing the same |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100792714B1 (en) | 2006-10-16 | 2008-01-11 | 한국에너지기술연구원 | Method of rule-based fault detection and diagnosis in air-handling system with detailed classification |
-
2015
- 2015-09-14 KR KR1020150129632A patent/KR20170031985A/en active Search and Examination
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100792714B1 (en) | 2006-10-16 | 2008-01-11 | 한국에너지기술연구원 | Method of rule-based fault detection and diagnosis in air-handling system with detailed classification |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20180130294A (en) * | 2017-05-29 | 2018-12-07 | 부경대학교 산학협력단 | Method for diagnosing machine fault based on sound |
| KR20190077358A (en) * | 2017-05-31 | 2019-07-03 | 신와 콘트롤즈 가부시키가이샤 | Status monitoring device, status monitoring method and program |
| US11405231B2 (en) | 2017-10-18 | 2022-08-02 | Samsung Electronics Co., Ltd. | Data learning server, and method for generating and using learning model thereof |
| WO2019078515A1 (en) * | 2017-10-18 | 2019-04-25 | 삼성전자주식회사 | Data learning server, and method for generating and using learning model thereof |
| KR20190043258A (en) * | 2017-10-18 | 2019-04-26 | 삼성전자주식회사 | Data learning server, method for generating and using thereof |
| KR20190071503A (en) * | 2017-12-14 | 2019-06-24 | 한전케이디엔주식회사 | Enhanced learning based energy usage virtual data generation system |
| CN108334999A (en) * | 2018-05-09 | 2018-07-27 | 山东交通学院 | The failure prediction method and system of fume hot-water type BrLi chiller |
| CN108334999B (en) * | 2018-05-09 | 2024-02-27 | 山东交通学院 | Fault prediction method and system for flue gas hot water type lithium bromide refrigerating unit |
| CN108647470A (en) * | 2018-05-29 | 2018-10-12 | 杭州电子科技大学 | A kind of localization method at the beginning of based on the leakage loss with depth belief network is clustered |
| CN109882996A (en) * | 2019-01-25 | 2019-06-14 | 珠海格力电器股份有限公司 | A kind of method and apparatus of control |
| KR20210061517A (en) * | 2019-11-19 | 2021-05-28 | 한국생산기술연구원 | Apparatus and method for fault diagnosis using fake data generated by machine learning |
| KR20210097412A (en) * | 2020-01-30 | 2021-08-09 | 전남대학교산학협력단 | Building Energy Failure Diagnosis and Analysis System with integrated Virtual sensor and Deep learning, and Building Energy Failure Diagnosis and Analysis Method |
| KR102268733B1 (en) * | 2020-06-02 | 2021-06-23 | 한국해양대학교 산학협력단 | Ship engine failure detection method and system |
| KR20220082597A (en) | 2020-12-10 | 2022-06-17 | 인천대학교 산학협력단 | Method and apparatus for sensing an energy system in a building |
| KR102281640B1 (en) * | 2021-03-24 | 2021-07-26 | 주식회사 유한테크 | AI Gas Leak Detection System with Self-Diagnosis Function and operating Method thereof |
| KR20230066874A (en) * | 2021-11-08 | 2023-05-16 | 중앙대학교 산학협력단 | Real-time self-adaptive steady state diagnosis method of heat pump system, recording medium and steady state diagnosis device for performing the same |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| KR20170031985A (en) | Fault detection and diagnostics method of air-conditioning system | |
| CN107111286B (en) | Automated functional testing for diagnostics and control | |
| RU2686252C2 (en) | Method of estimating normal or abnormal value of measured value of physical parameter of aircraft engine | |
| JP2021064370A (en) | Method and system for semi-supervised deep abnormality detection for large-scale industrial monitoring system based on time-series data utilizing digital twin simulation data | |
| JP6141235B2 (en) | How to detect anomalies in time series data | |
| EP2068220B1 (en) | methods and systems for performing diagnostics regarding underlying root causes in turbine engines | |
| US10539380B2 (en) | Method and system for thermographic analysis | |
| WO2016086360A1 (en) | Wind farm condition monitoring method and system | |
| ATE450830T1 (en) | SENSOR FAULT DIAGNOSIS AND PREDICTION USING A COMPONENT MODEL AND TIMESCALE ORTHOGONAL DEVELOPMENTS | |
| JP7340265B2 (en) | Abnormality detection device, abnormality detection method, and program | |
| EP1630635A2 (en) | Method and apparatus for improved fault detection in power generation equipment | |
| KR101797400B1 (en) | Method and apparatus for diagnosing fault based on probabilistic density | |
| CN111581763B (en) | Method for evaluating air path fault diagnosis result of aero-engine | |
| CN115277464A (en) | Cloud network change flow anomaly detection method based on multi-dimensional time series analysis | |
| CN103608815B (en) | Method and diagnostic system for supporting the controlled fault detection in technical systems | |
| US11567489B2 (en) | Method and device for efficient fault analysis through simulated faults in a digital twin | |
| CN112254274A (en) | Air conditioner fault recognition system based on machine learning technology | |
| US20220147034A1 (en) | Automated refinement of a labeled window of time series data | |
| KR20220049746A (en) | System for determining thermal abnormalities on exterior walls of building using thermal image and real image and method thereof | |
| CN110727669B (en) | Electric power system sensor data cleaning device and cleaning method | |
| CN105090000B (en) | A kind of method for diagnosing faults of the thick pumped vacuum systems of space simulator | |
| CN104460337B (en) | A kind of control system analysis of common cause failure method based on the amendment β factors | |
| US11339763B2 (en) | Method for windmill farm monitoring | |
| CN113919225B (en) | Environmental test box reliability assessment method and system | |
| CN111291239B (en) | Method, device, equipment and storage medium for testing unit |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| A201 | Request for examination | ||
| E902 | Notification of reason for refusal | ||
| AMND | Amendment | ||
| E902 | Notification of reason for refusal | ||
| AMND | Amendment | ||
| AMND | Amendment |