Industrial boiler monitoring system for artificial intelligence-based exhaust gas analysis and fault diagnosis

Abstract

The present invention relates to a monitoring system for industrial boilers for artificial intelligence-based exhaust gas analysis and fault diagnosis, and more specifically, to a method comprising: (a) a step of acquiring input / output data for components and input / output data for measured exhaust gas concentrations by operating the individual boiler of a customer for a predetermined period after commissioning; (b) a step of constructing big data for the individual boiler based on the acquired data; (c) a step of constructing a virtual sensor module for measuring virtual exhaust gas concentration and a fault diagnosis module for diagnosing component faults through machine learning techniques based on the big data; (d) a step of installing the virtual sensor module and the fault diagnosis module constructed in step (c) on the individual boiler and transmitting operation data resulting from the operation of the individual boiler to a management server; (e) a step in which the individual boiler calculates the exhaust gas concentration using the operation data via the virtual sensor module and calculates the probability of component failure using the fault diagnosis module. and (f) a step in which the calculated exhaust gas concentration and the diagnosis status of component failures in individual boilers are transmitted to and stored by a management server via an app and web, and the management server checks the combustion status of individual boilers in real time and issues an alarm when the boiler is operating abnormally due to incomplete combustion or component failure; is included.

Description

Technology Field

[0001] The present invention relates to an AI-based monitoring system for exhaust gas analysis and fault diagnosis of industrial boilers, which constructs a customized AI-based learning model for a customer to diagnose exhaust gas concentration and fault status of various components for individual boilers of a customer, and performs real-time diagnosis of exhaust gas concentration and faults of various components through the constructed learning model. Background Technology

[0002] Industrial boilers are the largest energy-consuming facilities in Korea, accounting for 30% of total energy consumption and 47% of industrial energy consumption, making the development of technologies for greenhouse gas reduction and energy saving essential. Accordingly, there is growing interest in real-time exhaust gas and fault diagnosis analysis devices specifically designed for industrial boilers that are affordable yet highly reliable.

[0003] If the emission concentration of substances is identified in real time, problems can be immediately detected when a problem occurs during boiler operation, and the air-fuel ratio can be re-set, thereby enabling a proactive response to air pollution caused by reduced efficiency.

[0004] The government is gradually expanding the installation of Continuous Emission Monitoring Systems (TMS) on combustion facilities at large-scale industrial sites to curb the emission of fine dust and air pollutants.

[0005] As of 2018, the total number of registered workplaces with combustion facilities in Korea is 56,151, but the number of workplaces with TMS installed is only 5,510, which is an absolute shortage.

[0006] Although the law targets only large-scale workplaces, the reality is that it is difficult for small businesses to purchase and operate TMS because the equipment itself is expensive, costing over hundreds of millions of won.

[0007] Consequently, while most industrial boilers are operated according to the air-fuel ratio set by the manufacturer during commissioning, there is a problem in that the amount of air pollutants generated and emitted in the exhaust gas is completely unknown until the boiler breaks down. Prior art literature

[0008] Korean Patent Publication No. 10-2078242 (February 11, 2020) The problem to be solved

[0009] The present invention has been devised to solve the problems described above.

[0010] One objective is to build a customized AI-based learning model for the client to diagnose exhaust gas concentrations and component failures in individual boilers, and to perform real-time diagnosis of exhaust gas concentrations and component failures through the built learning model.

[0011] Another objective of the present invention is to manufacture a real-time exhaust gas concentration measuring device at a lower cost compared to the purchase price of an existing Continuous Emission Monitoring System (TMS) by constructing a virtual sensor module for predicting exhaust gas concentration.

[0012] Another objective of the present invention is to enable immediate response via alarms by verifying the failure of critical boiler components in real time through the construction of a fault diagnosis module for diagnosing component failures, and to take preemptive measures by identifying component abnormalities in the event of incomplete combustion through real-time monitoring of exhaust gas concentration. means of solving the problem

[0014] A monitoring system for an industrial boiler for AI-based exhaust gas analysis and fault diagnosis according to the present invention comprises: (a) a step in which a service provider operates an individual boiler of a customer company for a predetermined period after commissioning to acquire input / output data for components and input / output data for measured exhaust gas concentrations; (b) a step of constructing big data for the individual boiler based on the acquired data; (c) a step of constructing a virtual sensor module for measuring virtual exhaust gas concentration and a fault diagnosis module for diagnosing component faults through machine learning techniques based on the big data; (d) a step of installing the virtual sensor module and the fault diagnosis module constructed in step (c) on the individual boiler of the customer company and transmitting operation data resulting from the operation of the individual boiler to a management server; (e) a step in which the individual boiler calculates the exhaust gas concentration using the operation data via the virtual sensor module and calculates the probability of component failure using the fault diagnosis module. and (f) a step in which the individual boiler transmits and stores the calculated exhaust gas concentration and the component fault diagnosis status to a management server of the app and web, and the service provider checks the combustion status of the individual boiler in real time and issues an alarm in the event of abnormal operation of the boiler due to incomplete combustion or component failure; is included.

[0015] The fault diagnosis module according to the present invention is characterized by being constructed using a Principal Component Analysis (PCA) model as a machine learning technique.

[0016] The fault diagnosis by the fault diagnosis module according to the present invention is characterized by being performed by a Reconstruction Based Contribution (RBC) technique based on real-time measurement values ​​of temperature, pressure, flow rate, etc., measured by a sensor mounted on a component, and a prediction value learned by the principal component analysis model.

[0017] The present invention comprises: (e-1) a step of detecting a sensor failure using the above-described abnormal contribution method; (e-2) a step of identifying a faulty sensor after detecting the sensor failure; (e-3) a step of performing an oversight correction for the faulty sensor; and (e-4) a step of detecting a sensor failure after the oversight correction; wherein, when a sensor failure is detected after the oversight correction, an alarm is emitted according to step (f) when a sensor failure is detected.

[0018] The virtual sensor module according to the present invention is characterized by being constructed using a regression model utilizing a statistically based PLS technique and an Auto-Encoder Neural Network as a machine learning technique.

[0019] The virtual sensor module according to the present invention is characterized by predicting oxygen (O2) concentration, sulfur oxide (SOx) concentration, nitrogen oxide (NOx) concentration, and carbon monoxide (CO) concentration by utilizing operation data of individual boilers. Effects of the invention

[0020] The monitoring system for industrial boilers for AI-based exhaust gas analysis and fault diagnosis according to the present invention constructs a customized AI-based learning model for each customer to diagnose the exhaust gas concentration and the failure status of various components for individual boilers of the customer, and can perform real-time diagnosis of exhaust gas concentration and faults of various components through the constructed learning model.

[0021] In addition, the present invention can be manufactured at a lower cost compared to the purchase price of an existing Continuous Emission Monitoring System (TMS) by constructing a virtual sensor module for predicting exhaust gas concentration, thereby ensuring economic feasibility.

[0022] In addition, the present invention enables real-time verification of failures in critical boiler components through the construction of a fault diagnosis module for diagnosing component failures, allowing for immediate response via alarms. Furthermore, it enables the identification of potential component failures related to incomplete combustion and, in the event of incomplete combustion, the identification of component abnormalities to take preemptive measures.

[0023] In addition, the present invention can achieve the optimization of the ratio of combustion to air in a boiler by actively preparing for combustion failures caused by excess air and non-ignition by measuring the oxygen concentration and carbon monoxide concentration in the exhaust gas in real time.

[0024] In addition, the present invention can reduce fuel costs by detecting excess air caused by abnormal oxygen concentration at an early stage, and can maximize the effect of preventive maintenance by preemptively identifying whether parts fail due to abnormal boiler air-fuel ratios. Brief explanation of the drawing

[0025] FIG. 1 is a flowchart showing a monitoring system for an industrial boiler for artificial intelligence-based exhaust gas analysis and fault diagnosis according to the present invention. FIG. 2 is a flowchart illustrating a fault diagnosis process in a monitoring system for an industrial boiler according to the present invention. FIG. 3 is a conceptual diagram showing a monitoring system for an industrial boiler according to the present invention. FIG. 4 is a process diagram and sensor location diagram of a target boiler used for accuracy verification and performance certification of an analysis model according to the present invention, do 5 is a drawing showing the monitoring results by actually applying the motor ringing system according to the present invention. Specific details for implementing the invention

[0026] In order to explain the present invention, the operational advantages of the present invention, and the objectives achieved by the implementation of the present invention, preferred embodiments of the present invention are illustrated below and examined with reference thereto.

[0027] First, the terms used in this application are used merely to describe specific embodiments and are not intended to limit the invention; singular expressions may include plural expressions unless the context clearly indicates otherwise. Furthermore, in this application, terms such as "comprising" or "having" are intended to specify the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0028] In describing the present invention, if it is determined that a detailed description of related known components or functions may obscure the essence of the invention, such detailed description is omitted.

[0029] As illustrated in FIGS. 1 to 3, the monitoring system for an industrial boiler for artificial intelligence-based exhaust gas analysis and fault diagnosis according to the present invention comprises: (a) a step of acquiring data for individual boilers (10) of a customer (S100); (b) a step of configuring big data (S200); (c) a step of constructing a virtual sensor module (40) and a fault diagnosis module (50) using machine learning techniques (S300); (d) a step of installing and operating the virtual sensor module (40) and the fault diagnosis module (50) constructed in the individual boiler (10) and transmitting the operation data to a management server (30) (S400); (e) a step of calculating the exhaust gas concentration and the possibility of component failure (S500); and (f) a step of checking the combustion state of the individual boiler (10) in real time and issuing an alarm when the boiler is operating abnormally due to incomplete combustion or component failure (S600).

[0030] First, as illustrated in FIGS. 1 and 3, step (a) (S100) according to the present invention is a process for a service provider (20) to operate a customer’s individual boiler (10) for a predetermined period after commissioning to obtain input / output data for the main parts of the boiler and input / output data for the measured exhaust gas concentration.

[0031] In particular, in step (a), in order to build a customized analysis model for each customer, individual boilers (10) of the customer are operated for about 2 to 3 weeks, and input / output data is collected and obtained. The data obtained in this way is used as basic data for building the virtual sensor module (40) and fault diagnosis module (50) to be described later.

[0032] Next, as illustrated in FIGS. 1 and 3, step (b) (S200) according to the present invention is a process for configuring big data for individual boilers (10) of a customer using the data obtained in step (a) (S100).

[0033] That is, in step (b) (S200), the data obtained from the individual boilers (10) of the customer companies are stored in the database (31) of the management server (30) for each customer company to be converted into big data, and a virtual sensor module (40) and a fault diagnosis module (50) are constructed through a machine learning technique utilizing this data.

[0034] As illustrated in FIGS. 1 and 3, step (c) (S300) according to the present invention is a process for custom-building a virtual sensor module (40) for measuring the concentration of virtual exhaust gas and a fault diagnosis module (50) for diagnosing component faults for each customer through machine learning techniques based on big data for each customer.

[0035] First, the virtual sensor module (40) can be constructed as a regression model using the statistically based Partial Least Squares (PLS) technique and the Auto-Encoder Neural Network among machine learning techniques.

[0036] In this case, since data for factory equipment such as boilers exhibits high collinearity among variables, Multivariate Statistics (PLS)—a machine learning technique based on multivariate statistics that extracts latent variables to reduce data dimensionality and interpret data characteristics—excellently performs well. Therefore, it is advisable to utilize PLS as the primary learning technique and Auto-Encoder Neural Networks as a secondary learning technique.

[0037] The virtual sensor module (40) configured in this way can measure oxygen (O2) concentration, sulfur oxide (SOx) concentration, nitrogen oxide (NOx) concentration and carbon monoxide (CO) concentration in real time. This utilizes operation data measured in real time from individual boilers (10), and for this purpose, various sensors for data collection can be placed in the individual boilers (10) of the customer company.

[0038] The city in Fig. 4 is a process diagram and sensor location diagram of a boiler used for verifying the accuracy and certifying the performance of the virtual sensor module (40).

[0039] Next, the fault diagnosis module (50) was constructed using a machine learning technique and a Principal Component Analysis (PCA) model.

[0040] In addition, the diagnosis of a fault is performed by the Reconstruction Based Contribution (RBC) technique based on real-time measurements of temperature, pressure, and flow rate from sensors mounted on important parts of individual boilers (10) and prediction values ​​learned by a principal component analysis model.

[0041] Fault diagnosis by the fault diagnosis module (50) is performed through sensor fault detection, fault sensor identification, and error correction.

[0042] Even if the fault diagnosis module (50) is accurate in detecting sensor failures, if the sensors, such as temperature, pressure, and flow rate, which are input variables of the model, are faulty, the predicted value will also have an error. Therefore, it is essential for the fault diagnosis module (50) to identify the failure of the input sensor and to provide an accurate predicted value instead of the measured value when the failure occurs.

[0043] To this end, an abnormality contribution technique based on oversight correction is used to determine whether the sensor is functioning normally.

[0044] RBC is a method for identifying faulty sensors that assumes all sensors are faulty, sequentially calculates the error magnitude for each sensor using a PCA model, and then identifies the faulty sensor based on the error magnitude.

[0045] As described above, the diagnosis of a fault in a sensor by the fault diagnosis module (50) according to the present invention is performed through a process such as sensor fault detection, fault sensor identification, and error correction, and the specific process for this will be described later.

[0046] As illustrated in FIGS. 1 and 3, step (d) (S400) according to the present invention is a process for installing a constructed virtual sensor module (40) and a fault diagnosis module (50) in an individual boiler (10) and transmitting operation data according to the operation of the individual boiler (10) to a management server (30).

[0047] (d) In step (S400), the service provider establishes the virtual sensor module (40) and the fault diagnosis module (50) as described above, and when the virtual sensor module (40) and the fault diagnosis module established in this way are distributed to the customer through online / offline channels, the customer installs and applies the virtual sensor module (40) and the fault diagnosis module (50) to individual boilers (10) so as to be linked with the boiler.

[0048] In addition, the client company transmits the operation data generated through the operation of individual boilers (10) to the management server (30) via a wired or wireless communication network or the internet network, and the management server (30) stores the operation data in real time in a database (31) to create big data.

[0049] As illustrated in FIGS. 1 to 3, step (e) (S500) according to the present invention is a process for calculating the exhaust gas concentration by the virtual sensor module (40) using operation data in the individual boiler (10) and calculating the possibility of a component failure by the fault diagnosis module (50).

[0050] That is, in step (e) (S500), the oxygen (CO2) concentration, sulfur oxide (SOx) concentration, nitrogen oxide (NOx) concentration and carbon monoxide (CO) concentration are measured in real time by the virtual sensor module (40).

[0051] In particular, since the ratio of combustion to air is paramount for optimal boiler combustion, measuring O2 and CO concentrations in the exhaust gas in real time allows for proactive measures against combustion failures caused by excess air or ignition issues. Furthermore, the ability to detect excess air caused by abnormal O2 concentrations at an early stage enables fuel cost savings. Additionally, the ability to preemptively identify component issues resulting from abnormal boiler air-fuel ratios maximizes the effectiveness of preventive maintenance.

[0052] In addition, as shown in FIG. 2, the sensor failure of the main component is diagnosed by the fault diagnosis module (50), and the process is as follows.

[0053] In step (e-1) (S510), a sensor failure is detected using the abnormal contribution method, and in step (e-2) (S520), the faulty sensor is identified after the detection of the sensor failure. Additionally, in step (e-3) (S530), a false positive correction is performed on the faulty sensor, and in step (e-4) (S540), the sensor failure is detected after the false positive correction. In this case, when the sensor failure is detected after the false positive correction, if the sensor failure is detected, an alarm is triggered by step (f) (S600), which will be described later, so that the client or field manager recognizes it.

[0054] Specific operations in each process for fault diagnosis can be performed as described above.

[0055] In this case, regarding the status of measurement sensor failure, if no sensor failure is detected during the sensor failure detection stage, all sensor values ​​are normal.

[0056] Conversely, if a sensor failure is detected during the sensor failure detection stage, the faulty sensor is identified using anomaly contribution based on over-correction. After performing over-correction on the faulty sensor, if the failure is resolved when determining whether the sensor is faulty, it is diagnosed as a single sensor failure among the measured values. In this case, if the sensor failure is not resolved when determining whether the sensor is faulty after performing over-correction, multiple sensors are diagnosed as being in a faulty or process abnormal state.

[0057] In cases where a sensor failure is diagnosed as such, the service provider transmits an alarm to the client or on-site manager via an app or web through real-time monitoring to enable immediate response.

[0058] As illustrated in FIGS. 1 and 3, step (f) (S600) according to the present invention is a process for transmitting and storing the calculated exhaust gas concentration and component fault diagnosis status from an individual boiler (10) to a management server (30) via an app and web, and for a service provider (20) to check the combustion status of the individual boiler (10) in real time and to perform an alarm when the boiler is operating abnormally due to incomplete combustion and component failure.

[0059] That is, in step (f) (S600), the service provider monitors in real time the exhaust gas concentration and component fault diagnosis results calculated from the boilers installed at the site, i.e., individual boilers (10) installed at multiple customer companies, and transmits an alarm to the customer company or site manager via an app installed on a terminal, or emits an alarm using a communication network or the internet network.

[0060] Upon receiving such an alarm, it is advisable for the client or site manager to take prompt action, such as stopping the boiler's operation or replacing or repairing faulty parts, to prevent a decrease in boiler efficiency.

[0061] The monitoring system according to the present invention includes an individual boiler (10) of a customer company, a virtual sensor module (40) and a fault diagnosis module (50) installed in the individual boiler (10). In addition, the individual boiler (10) of the customer company, the virtual sensor module (40), and the fault diagnosis module (50) are configured to transmit and receive data from a management server (30), and a service provider (20) monitors the exhaust gas concentration and fault diagnosis results for the individual boilers (10) of the customer company in real time through a wired or wireless communication network or an internet network, and in the event of an abnormal exhaust gas concentration or a malfunction of a sensor, transmits an alarm to the customer company or a field manager in real time to make them aware of it.

[0062] The city in Figure 5 shows the results monitored by the service provider.

[0063] In Fig. 5, the red window confirms incomplete combustion through the O2 and CO values ​​of the virtual exhaust gas, for example, when incomplete combustion occurs due to a blower inverter failure, and identifies whether the blower inverter is faulty through the fault diagnosis module and provides guidance.

[0064] In Figure 5, the yellow window shows that the emission concentrations of nitrogen oxides (NOx) and carbon dioxide (CO2) can be identified in real time through AI-based exhaust gas analysis, and the emission amount calculated according to the formula can be identified.

[0065] As such, the present invention has been described with reference to an embodiment illustrated in the drawings, but this is merely illustrative, and those skilled in the art will understand that various modifications and equivalent alternative embodiments are possible therefrom.

[0066] Therefore, the true scope of technical protection of the present invention should be determined by the technical concept of the appended claims. Explanation of the symbols

[0067] 10: Individual boiler 20 : Service provider 30 : Management Server 31 : Database 40 : Virtual sensor module 50 : Fault diagnosis module

Claims

Claim 1 (a) A service provider (20) obtains input / output data for parts and input / output data for measured exhaust gas concentration by operating the individual boiler (10) of a customer company for a predetermined period after commissioning (S100); (b) constructs big data for the individual boiler (10) based on the obtained data (S200); (c) constructs a virtual sensor module (40) for measuring virtual exhaust gas concentration and a fault diagnosis module (50) for diagnosing part faults using machine learning techniques based on the big data (S300); (d) installs the virtual sensor module (40) and the fault diagnosis module (50) constructed in step (c) on the individual boiler (10) of the customer company and transmits operation data resulting from the operation of the individual boiler (10) to a management server (30) (S400); (e) the individual boiler (10) calculates the exhaust gas concentration using the operation data through the virtual sensor module (40), and the A step (S500) of calculating the possibility of a component failure by a fault diagnosis module (50); and (f) the individual boiler (10) transmits and stores the calculated exhaust gas concentration and the component failure diagnosis status to a management server (30) via an app and web, and the service provider (20) checks the combustion status of the individual boiler (10) in real time and issues an alarm when the boiler is operating abnormally due to incomplete combustion or component failure (S600);The above fault diagnosis module (50) is constructed using a machine learning technique with a Principal Component Analysis (PCA) model, and the fault diagnosis by the above fault diagnosis module (50) is performed by a Reconstruction Based Contribution (RBC) technique based on real-time measurement values ​​of temperature, pressure, and flow rate measured by sensors mounted on the parts and prediction values ​​learned by the above Principal Component Analysis model, and the above virtual sensor module (40) is constructed using a machine learning technique with a regression model utilizing a statistical-based PLS technique and an Auto-Encoder Neural Network, and the above virtual sensor module (40) predicts oxygen (O2) concentration, sulfur oxide (SOx) concentration, nitrogen oxide (NOx) concentration, and carbon monoxide (CO) concentration using operation data of individual boilers (10), and the virtual sensor module (40) of step (c) is, in order to reduce the dimensionality of equipment data with high collinearity between variables, the A monitoring system for an industrial boiler for AI-based exhaust gas analysis and fault diagnosis, wherein the system is constructed as a regression model that combines the Partial Least Squares (PLS) method as a learning technique and an Auto-Encoder Neural Network as an auxiliary learning technique, and the fault diagnosis module (50) of step (e) is performed using a Principal Component Analysis (PCA) model and an abnormality contribution (RBC) method based on an abnormality correction, and includes (e-1) a process of identifying a faulty sensor when a sensor fault is detected, (e-2) a process of performing an abnormality correction on the identified faulty sensor using a predicted value, and (e-3) a process of re-detecting whether the sensor is faulty after the abnormality correction, and if the sensor fault alarm is released as a result of the re-detection, it is diagnosed as a single sensor fault, and if the alarm is not released, it is diagnosed as a multi-sensor fault or a process abnormality state and classified. Claim 2 delete Claim 3 delete Claim 4 delete Claim 5 delete Claim 6 delete

Images