JP7730726B2 - Heat transfer surface cleaning system - Google Patents

Description

This disclosure relates to a heat transfer surface cleaning system that cleans the heat transfer surfaces of heat transfer tubes installed in a boiler and removes deposits from the heat transfer surfaces.

In waste heat boilers or coal-fired boilers that use high-temperature exhaust gas from waste incinerators, the exhaust gas inside the furnace contains a high concentration of dust. The furnace walls of such boilers are constructed with connected heat transfer tubes through which cooling water flows. When such dust-rich exhaust gas is introduced into the boiler, the dust gradually adheres to the heat transfer surfaces of the heat transfer tubes.

As the thickness of deposits on the heat transfer surface increases, the amount of heat exchanged between the furnace interior and the heat transfer tubes and cooling water decreases. There is also the risk of the deposits corroding the heat transfer surface. For this reason, regular inspections and cleaning work have traditionally been carried out.

JP 2019-105393 A Japanese Unexamined Patent Publication No. 62-299612

A known configuration is to clean the heat transfer surfaces inside a boiler by spraying water onto them, etc. However, the amount of deposits that adhere to the heat transfer surfaces inside a boiler is not uniform across the entire surface. If the entire heat transfer surface is cleaned based on the deposit status on only a portion of the heat transfer surface inside the boiler, the cleaning pressure may be excessive in areas with a small amount of deposits, which could result in the heat transfer surface being scraped away. On the other hand, there is a risk that the heat transfer surface may not be cleaned sufficiently in areas with a large amount of deposits.

In this regard, the above-mentioned Patent Document 1 discloses measuring the inlet steam temperature, outlet steam temperature, inlet exhaust gas temperature, and outlet exhaust gas temperature of a heat transfer tube, estimating the thickness of dust adhering to the heat transfer tube from these values, and continuing cleaning until the thickness estimated from each value falls below a threshold. However, the configuration of Patent Document 1 requires cleaning to be performed at a fixed cleaning capacity for a specified cleaning time, and then determining whether the thickness has fallen below the threshold. For this reason, in order to prevent excessive cleaning, the cleaning capacity for the specified cleaning time must be set low in advance, and if the thickness is large, it will take a long time to complete cleaning.

Patent Document 2 also discloses a configuration in which, after performing overall soot blowing, the temperature difference between the inlet and outlet of an economizer, etc., the feedwater temperature difference, or the above-mentioned temperature differences are measured and compared to determine whether or not these differences have become larger than a predetermined value, thereby determining whether or not to perform further individual soot blowing. However, like Patent Document 1, the configuration of Patent Document 2 only performs feedback control, and it is not clear what level of cleaning power is required to clean the thickness of deposits adhering to the heat transfer tubes.

This disclosure has been made to solve the above-mentioned problems, and aims to provide a heat transfer surface cleaning system that can apply appropriate cleaning capabilities to the thickness of deposits attached to the heat transfer surface.

A heat transfer surface cleaning system according to one aspect of the present invention includes an infrared camera that measures the surface temperature of a heat transfer surface exposed inside a boiler by photographing the heat transfer surface; a thermometer that measures the gas temperature in the boiler's internal space; a cleaning device that cleans the heat transfer surface by spraying a cleaning fluid; and a controller that controls the amount of cleaning fluid sprayed. The controller outputs the spray amount, acquires state values including the measured surface temperature distribution of the heat transfer surface and the gas temperature, acquires the post-cleaning gas temperature measured by the thermometer after cleaning by the cleaning device based on the spray amount, calculates a reward based on the post-cleaning gas temperature, and performs machine learning to determine the injection amount to be output for the current state value based on the injection amount, the state quantity, and the reward.

This disclosure makes it possible to apply appropriate cleaning capabilities to the thickness of deposits adhering to the heat transfer surface.

FIG. 1 is a schematic cross-sectional view showing the internal structure of an incineration plant to which a heat transfer surface monitoring device and a heat transfer surface cleaning system according to this embodiment are applied. FIG. 2 is a cross-sectional view of the radiation chamber shown in FIG. 1 taken along line II-II. FIG. 3 is a diagram showing the infrared camera shown in FIG. 2 when rotated. FIG. 4 is a block diagram showing a schematic configuration of a heat transfer surface monitoring device according to this embodiment. FIG. 5 is a block diagram showing a schematic configuration of a heat transfer surface cleaning system according to this embodiment. FIG. 6 is a flowchart showing the flow of machine learning performed by the controller shown in FIG.

The following describes a heat transfer surface monitoring device and a heat transfer surface cleaning system using the same according to an embodiment of the present disclosure, with reference to the drawings. This embodiment illustrates an example in which the heat transfer surface monitoring device and the heat transfer surface cleaning system are applied to an incineration plant with a waste heat boiler.

[Heat transfer surface monitoring device]
First, a heat transfer surface monitoring device according to this embodiment will be described. Fig. 1 is a schematic cross-sectional view showing the internal structure of an incineration plant to which the heat transfer surface monitoring device and heat transfer surface cleaning system according to this embodiment are applied. The incineration plant 100 shown in Fig. 1 includes an incinerator 102 having a furnace chamber 103 for incinerating waste using an oxygen-containing gas, and a boiler 104, which is a steam recovery device that recovers exhaust heat as steam from the combustion exhaust gas discharged from the incinerator 102. A hopper 105 and a chute 106 are disposed on the opposite side of the incinerator 102 from the boiler 104, i.e., on the upstream side. An exhaust path 107 for the combustion exhaust gas extends downstream from the boiler 104 to a chimney 108. For example, an economizer, a temperature reducing tower, a dust collector, and a blower may be provided in the exhaust path 107, in this order from the upstream side.

The incinerator 102 has a stoker installed below the furnace chamber 103. The stoker functions as a means of transporting waste. The stoker comprises, in order from the side closest to the chute 106, a drying stoker 111, a combustion stoker 112, and a post-combustion stoker 113. That is, these stokers 111, 112, and 113 are arranged in the direction of waste movement. Wind boxes 114, 115, and 116 are installed below the drying, combustion, and post-combustion stokers 111, 112, and 113, respectively.

Furthermore, the incinerator 102 has a re-burning chamber 117 that is continuous with the furnace chamber 103 between the furnace chamber 103 and the boiler 104. While the combustion stoker 112 has one stage in the illustrated example, it may have two or more stages. Each stoker 111, 112, 113 operates intermittently, for example, at different intervals.

In the furnace chamber 103, combustion gas is generated by the thermal decomposition and partial oxidation of the waste, and this combustion gas is burned together with the waste. The re-burning chamber 117 is used to completely burn the combustion gas that flows out of the furnace chamber 103. The dust remaining after the waste combustion is discharged from an outlet 118 located adjacent to the post-burning stoker 113.

In the boiler 104, steam is generated using waste heat from the combustion exhaust gas discharged from the incinerator 102. More specifically, as shown in FIG. 1, the boiler 104 is equipped with an exhaust gas passage 109 through which the combustion exhaust gas passes. The exhaust gas passage 109 includes a radiant chamber 119 located above the re-burning chamber 117, a first flue 120 whose upper section communicates with the radiant chamber 119, and a second flue 121 whose lower section communicates with the first flue 120. In other words, the boiler 104 has an internal space consisting of the radiant chamber 119, the first flue 120, and the second flue 121.

A plurality of heat transfer tubes 123 are provided in each of the walls defining the radiant chamber 119 and the first and second flue 120 and 121. The heat transfer tubes 123 are made of carbon steel, such as STB340. The boiler 104 is equipped with a boiler drum 124 connected to the heat transfer tubes 123. Water delivered from the boiler drum 124 flows through the heat transfer tubes 123. The water in the heat transfer tubes 123 recovers waste heat from the radiant chamber 119 and the first flue 120, and some of it evaporates to become brackish water, which is then returned to the boiler drum 124. Some of the brackish water returned to the boiler drum 124 vaporizes to become steam.

A superheater 125 is provided in the second flue 121. The superheater 125 has superheater tubes 126 for superheating the steam in the boiler drum 124 with the heat of combustion or the heat of the flue gas. The superheated steam, which has been superheated by the superheater 125 to a high temperature and pressure, is sent to a turbine 128 connected to a generator 127 and used to generate electricity. Most of the flue gas that passes through the boiler 104 flows through the exhaust path 107 and is then released into the atmosphere from the chimney 108.

Figure 2 is a cross-sectional view taken along line II-II of the radiation chamber shown in Figure 1. As shown in Figure 2, the radiation chamber 119 has a rectangular cross section defined by four heat transfer surfaces F1, F2, F3, and F4. Each of the four heat transfer surfaces F1, F2, F3, and F4 has a membrane wall to which multiple heat transfer tubes 123 are connected. Hereinafter, the four heat transfer surfaces will be referred to as the first heat transfer surface F1, the second heat transfer surface F2, the third heat transfer surface F3, and the fourth heat transfer surface F4. The second heat transfer surface F2 faces the first heat transfer surface F1, and the fourth heat transfer surface F4 faces the third heat transfer surface F3.

The first heat transfer surface F1 and the second heat transfer surface F2 are planes parallel to a first imaginary plane that includes the arrangement direction of the radiation chamber 119, the first flue 120, and the second flue 121, i.e., planes parallel to the plane of the paper in Figure 1, and the third heat transfer surface F3 and the fourth heat transfer surface F4 are planes that intersect with the first heat transfer surface F1 and the second heat transfer surface F2. The third heat transfer surface F3 is also a heat transfer surface that defines the first flue 120.

The first heat transfer surface F1 has a first window W1 through which infrared rays pass. The heat transfer surface monitoring device 1 is equipped with a first infrared camera 11 located outside the first heat transfer surface F1 so as to photograph the second heat transfer surface F2 through the first window W1. Furthermore, the heat transfer surface monitoring device 1 is equipped with a second infrared camera 12 located outside the second heat transfer surface F2 so as to photograph the first heat transfer surface F1 through the second window W2. Position P1 in Figure 1 indicates the positions of the first window W1 and the second window W2. Position P1 is located, for example, in the center of the radiation chamber 119 in the direction of waste gas flow.

The first infrared camera 11 measures the surface temperature of the second heat transfer surface F2 by photographing the second heat transfer surface F2. Similarly, the second infrared camera 12 measures the surface temperature of the first heat transfer surface F1 by photographing the first heat transfer surface F1. The measurement results from the infrared cameras 11 and 12 are obtained as the surface temperature distribution within the imaging range of the infrared cameras 11 and 12. The surface temperature distribution is, for example, image data in which the imaging range is color-coded into multiple colors that differ according to a predetermined temperature range.

The heat transfer surface monitoring device 1 further includes a thermometer 13 that measures the gas temperature within the radiation chamber 119. The thermometer 13 includes, for example, a temperature sensor with a thermocouple. The thermometer 13 is disposed within the radiation chamber 119. The location of the thermometer 13 is not particularly limited, but it is disposed, for example, within the angle of view of the first infrared camera 11 or the second infrared camera 12.

The first infrared camera 11 is rotatable around a predetermined rotation axis R1 that is aligned in the vertical direction. The second infrared camera 12 is similarly rotatable around a predetermined rotation axis R2 that is aligned in the vertical direction. Both rotation axis R1 and rotation axis R2 are parallel to the first heat transfer surface F1 or the second heat transfer surface F2 and parallel to the third heat transfer surface F3 or the fourth heat transfer surface F4.

Figure 3 is a diagram showing the infrared camera shown in Figure 2 when rotated. The example in Figure 3 shows a state in which the first infrared camera 11 and the second infrared camera 12 are both rotated to face the third heat transfer surface F3. In this way, by rotating the first infrared camera 11 or the second infrared camera 12, it is possible to photograph the third heat transfer surface F3 and measure the surface temperature of the third heat transfer surface F3. Similarly, by rotating the first infrared camera 11 or the second infrared camera 12, it is possible to photograph the fourth heat transfer surface F4 and measure the surface temperature of the fourth heat transfer surface F4.

The angle of view of the first infrared camera 11 may be set, for example, so that it includes both horizontal ends of the second heat transfer surface F2 when the first infrared camera 11 is positioned directly opposite the second heat transfer surface F2. However, a narrower angle of view may also be used. In this case, the surface temperature distribution across the entire horizontal area of the second heat transfer surface F2 may be obtained by rotating the first infrared camera 11 by a predetermined angle around the rotation axis F1 as described above and taking multiple images. The same applies to the second infrared camera 12.

Figure 4 is a block diagram showing the schematic configuration of a heat transfer surface monitoring device in this embodiment. The heat transfer surface monitoring device 1 includes the first infrared camera 11, second infrared camera 12, thermometer 13, and computing unit 14 described above. The computing unit 14 is configured by a computer equipped with storage for storing various data. For example, the computing unit 14 includes a CPU, main memory (RAM), storage, a communication interface, etc.

It should be noted that the functions of the elements disclosed herein can be performed using circuits or processing circuits, including general-purpose processors, special-purpose processors, integrated circuits, ASICs (Application Specific Integrated Circuits), conventional circuits, or combinations thereof, configured or programmed to perform the disclosed functions. A processor is considered a processing circuit or circuit because it includes transistors and other circuitry. In this specification, a circuit, unit, or means (...part) is hardware that performs the recited functions or hardware that is programmed to perform the recited functions. The hardware may be hardware disclosed herein or other known hardware that is programmed or configured to perform the recited functions. Where the hardware is a processor, which is considered a type of circuit, the circuit, unit, or means is a combination of hardware and software, and the software is used to configure the hardware and/or processor.

Calculator 4 acquires the surface temperature distribution of the monitored heat transfer surfaces F1, F2, F3, and F4 from the corresponding infrared cameras 11 and 12. Furthermore, calculator 14 acquires the gas temperature of the internal space of boiler 104 partitioned by the monitored heat transfer surfaces F1, F2, F3, and F4 from thermometer 13. Calculator 4 performs an estimation calculation to estimate the thickness of deposits adhering to the corresponding heat transfer surfaces using the surface temperature and gas temperature of the heat transfer surfaces.

Here, the surface temperature Td of the dust deposits is expressed by the following formula using the gas temperature Tg and the dust thickness td:

Here, K is the overall heat transfer coefficient of the heat transfer tube 123, Kd is the overall heat transfer coefficient at the surface of the dust, and Tw is the temperature of the water flowing through the heat transfer tube 123. αc is the convection heat transfer coefficient, αr is the radiation heat transfer coefficient, and αw is the internal heat transfer coefficient of the heat transfer tube 123. λ is the thermal conductivity of the heat transfer tube 123, and t is the wall thickness of the heat transfer tube 123. λd is the thermal conductivity of the dust.

All values used in the above equations (1) to (4) other than the dust surface temperature Td, gas temperature Tg, and dust thickness td are given in advance from thermal calculations at the time of design or average measured values, etc. Therefore, the above equations (1) to (4) can be rewritten as follows to calculate the dust thickness td from the dust surface temperature Td and gas temperature Tg:

The dust surface temperature Td corresponds to the surface temperature of the heat transfer surface acquired by the infrared cameras 11 and 12. The gas temperature Tg is the temperature acquired by the thermometer 13. The calculator 14 acquires the gas temperature Tg measured by the thermometer 13 and also acquires the temperature at a certain position coordinate from the surface temperature distribution of the heat transfer surface, and uses the acquired temperature as the dust surface temperature Td to calculate the dust thickness td using the above equation (5). That is, the first term of the above equation (5) includes variables Tg and Td, and the second term is a constant.

In the above equation (5), as the gas temperature Tg increases, the numerator of the first term becomes larger than the denominator, and therefore the dust thickness td increases. This means that as the dust thickness td increases, heat exchange between the exhaust gas in the boiler 104 and the heat transfer tubes 123 decreases, thereby suppressing the decrease in gas temperature Tg. Furthermore, in the above equation (5), as the dust surface temperature Td increases, the denominator of the first term becomes smaller, and therefore the dust thickness td increases. This means that as the dust thickness td increases, heat exchange between the exhaust gas in the boiler 104 and the heat transfer tubes 123 decreases, thereby increasing the dust surface temperature Td. Therefore, the higher the gas temperature Tg, the larger the dust thickness td, and the higher the dust surface temperature Td, the larger the dust thickness td.

The calculator 14 calculates the surface temperature distribution of the heat transfer surface, i.e., the dust thickness td for multiple position coordinates in the images captured by the infrared cameras 11 and 12. For example, the calculator 14 acquires temperatures at predetermined coordinate position intervals from the surface temperature distribution and calculates the dust thickness td for each acquired temperature. This allows the dust thickness distribution on the heat transfer surface within the image capture range of the infrared cameras 11 and 12 to be obtained.

As described above, with the heat transfer surface monitoring device 1 of this embodiment, the surface temperatures of the heat transfer surfaces F1, F2, F3, and F4 obtained from the images captured by the infrared cameras 11 and 12, and the gas temperature inside the boiler 104, can be used to obtain the thickness td of the dust deposits as three-dimensional information in which thickness information is added to the planar images captured by the infrared cameras 11 and 12. Therefore, the thickness distribution of dust adhering to the heat transfer surfaces F1, F2, F3, and F4 can be estimated quantitatively and in real time. Furthermore, by repeatedly capturing images over time, it is also possible to obtain the temporal change in the thickness distribution of dust adhering to the heat transfer surfaces F1, F2, F3, and F4.

Furthermore, because the infrared cameras 11 and 12 are positioned opposite each other, it is possible to estimate the dust thickness distribution on both of the opposing heat transfer surfaces F1 and F2. Furthermore, because the infrared cameras 11 and 12 rotate around the rotation axes R1 and R2, it is also possible to estimate the dust thickness distribution on the heat transfer surfaces F3 and F4 that are perpendicular to the first heat transfer surface F1 and second heat transfer surface F2 on which the first window W1 and second window W2 for the infrared cameras 11 and 12 are installed. Therefore, it is possible to monitor the status of a wide range of heat transfer surfaces while minimizing the number of parts required for the heat transfer surface monitoring device 1.

The positions of the infrared cameras 11 and 12 can be set appropriately depending on the position of the heat transfer surface to be measured. For example, to obtain the dust thickness distribution on the heat transfer surface in the first flue 120, the infrared cameras 11 and 12 are installed at positions P2 and P3 in the first flue 120, or both. Furthermore, to obtain the dust thickness distribution on the heat transfer surface in the second flue 121, the infrared cameras 11 and 12 are installed at positions P4 and P5 in the second flue 121, or both.

In this way, infrared cameras 11, 12 can be installed at one or more locations in each exhaust gas path separated by heat transfer tubes 123. When infrared cameras 11, 12 are installed at multiple locations, they may be installed in the flow direction of the exhaust gas path, for example, in the vertical direction, or in a direction intersecting the flow direction of the exhaust gas path, for example, in the horizontal direction. Furthermore, in this embodiment, an example is shown in which infrared cameras 11, 12 are rotatable in the horizontal direction, but in addition to this, or instead, they may also be rotatable in the vertical direction. For example, infrared cameras 11, 12 may be rotatable around a rotation axis parallel to a horizontal plane.

By installing infrared cameras 11 and 12 in multiple locations, rotating infrared cameras 11 and 12, or combining these, it is possible to capture images of the entire desired heat transfer surface. In other words, the surface temperature distribution of the entire heat transfer surface can be obtained. Therefore, the calculator 14 can estimate the thickness distribution of dust adhering to the entire heat transfer surface from the surface temperature distribution of the entire heat transfer surface.

Thermometers 13 measuring the gas temperature within the boiler 104 may also be installed at multiple locations within the boiler 104. For example, when monitoring the heat transfer surface in the first flue 120, the thermometer 13 may be installed within the first flue 120. Furthermore, when monitoring the heat transfer surface in the second flue 121, the thermometer 13 may be installed within the second flue 121. Furthermore, thermometers 13 may be installed in each of the radiation chamber 119, the first flue 120, and the second flue 121. In this case, the gas temperature measured by the thermometer 13 within the radiation chamber 119 may be used to estimate the dust thickness distribution on the heat transfer surface of the radiation chamber 119, the gas temperature measured by the thermometer 13 within the first flue 120 may be used to estimate the dust thickness distribution in the first flue 120, and the gas temperature measured by the thermometer 13 within the second flue 121 may be used to estimate the dust thickness distribution in the second flue 121.

As described above, by obtaining the dust thickness distribution on the heat transfer surfaces F1, F2, F3, and F4, it is possible to set the maintenance period for cleaning the inside of the boiler 104 at an efficient time. For example, the calculator 14 may display the dust thickness distribution on a specified monitor as image data that is color-coded in multiple colors according to a specified thickness range. Furthermore, it is possible to pinpoint the locations within the boiler 104 where cleaning should be performed.

For example, maintenance may be performed when the area of the entire heat transfer surface where the dust thickness td exceeds a predetermined threshold reaches or exceeds a predetermined reference area. For example, the calculator 14 may display an image of the thickness distribution on a monitor and highlight the areas on the image where the dust thickness exceeds the threshold. Highlighting may include, for example, flashing the areas. Furthermore, the areas to be cleaned when performing maintenance may be within a predetermined range centered on the areas where the predetermined threshold is exceeded. Maintenance or cleaning work may be performed manually by an operator entering the boiler 104, or may be performed using the cleaning device 21 described below.

[Heat transfer surface cleaning system]
Furthermore, in this embodiment, the incineration plant 100 is equipped with a heat transfer surface cleaning system 2 that cleans the inside of the boiler 104. Fig. 5 is a block diagram showing a schematic configuration of the heat transfer surface cleaning system in this embodiment. As shown in Fig. 5, the heat transfer surface cleaning system 2 also includes infrared cameras 11 and 12 and a thermometer 13, similar to the heat transfer surface monitoring device 1. The infrared cameras 11 and 12 and the thermometer 13 may be shared by the heat transfer surface monitoring device 1 and the heat transfer surface cleaning system 2, or the infrared cameras 11 and 12 and the thermometer 13 for the heat transfer surface monitoring device 1 and the infrared cameras 11 and 12 and the thermometer 13 for the heat transfer surface cleaning system 2 may be installed in the boiler 104.

The heat transfer surface cleaning system 2 further includes a cleaning device 21 and a controller 22. The cleaning device 21 is not particularly limited, but may be, for example, a water jet cleaning device having a water jet nozzle 21a that sprays water as a cleaning fluid. The cleaning device 21 includes a main body 21b installed outside the boiler 104 and a water supply pipe 21c that extends into the boiler 104 and connects the main body 21b to the water jet nozzle 21a. The water supply pipe 21c is inserted into the boiler 104 through an insertion hole located in the ceiling of the boiler 104. The main body 21b includes a pump that pumps water. Water from the main body 21b flows through the water supply pipe 21c, and water is sprayed into the boiler 104 from the water jet nozzle 21a. The main body 21b includes a drive unit that changes the direction or range of water spray from the water jet nozzle 21a and changes the vertical position of the water jet nozzle 21a.

The controller 22 is configured as a computer equipped with storage for storing various data. For example, the controller 22 includes a CPU, main memory (RAM), storage, a communication interface, etc.

The controller 22 controls the operation of the cleaning device 21. For example, the controller 22 controls the vertical position of the water injection nozzle 21a. The controller 22 also controls the direction or range of water injection from the water injection nozzle 21a. This controls the position at which water is injected from the water injection nozzle 21a relative to the heat transfer surface. The controller 22 also controls the amount of water injected from the water injection nozzle 21a.

The controller 22 controls the position of the water injection nozzle 21a and the amount of water injected based on the dust thickness distribution estimated by the calculator 14 of the heat transfer surface monitoring device 1. For example, the controller 22 sets the target position based on the dust thickness distribution to a position where the dust thickness is equal to or greater than a predetermined threshold, and controls the position of the water injection nozzle 21a so that water is injected toward the target position. The controller 22 also controls the amount of water injected according to the dust thickness at the target position. The amount of water injected is adjusted, for example, by changing the injection time, changing the amount injected per unit time, or changing the number of injections when a predetermined amount is injected per injection.

Regarding control of the amount of water sprayed, the controller 22 may determine the amount of sprayed using a trained model based on machine learning. The following example illustrates a mode in which the controller 22 generates a trained model through reinforcement learning. In reinforcement learning, the agent, which is the subject of learning, progresses in learning through interactions between the agent and the environment to be controlled. In this embodiment, the agent is the controller 22, and the environment to be controlled is the boiler 104 and the cleaning device 21. More specifically, when the agent learns, the following steps (A) to (D) are repeated.
(A) An agent observes the state S _t of the environment at time T.
(B) The agent selects an action A _t that it can take based on observation results and past learning, and then executes the action A _t .
(C) When an action A _t is executed, the state S _t of the environment changes to the next state S _t+1 , and the agent receives a reward R _t+1 based on this change in state.
(D) The agent progresses in learning based on the state S _t , the action A _t , the reward R _t+1 and the results of past learning.

In the learning in (D) above, the agent acquires a mapping of state S _t , action A _t , and reward R _t+1 as reference information for determining the amount of reward R that can be obtained in the future. For example, if the number of states that can be taken at each time is m and the number of actions that can be taken is n, an m×n two-dimensional array is obtained that stores the reward R _t+1 for the pair of state S _t and action A _t by repeating the actions. Then, based on the obtained mapping, a value function or evaluation function that indicates how good the current state and action are is used, and the value function is updated while the action is repeated, thereby learning the optimal action for the state.

The state-action value function Q(S _t , A _t ), known as one of the value functions, indicates how good an action A _t is in a certain state S _t . The state-action value function Q(S _t , A _t ) is expressed as a function that takes the state and the action as arguments. In learning through repeated actions, the state-action value function Q(S _t , A _t ) is updated based on the reward obtained for an action in a certain state, the value of the action in a future state to which the action will lead, and the like. The update equation for the state-action value function Q(S _t , A _t ) is defined according to the reinforcement learning algorithm. For example, in Q-learning, which is one of the representative reinforcement learning algorithms, the update equation for the state-action value function Q(S _t , A _t ) is defined by the following equation (6):

however,
Q: State-action value function S _t : State at time t A _t : Action for the state at time t R _t+1 : Reward obtained from the state at time t+1 η: Learning coefficient (0<η≦1)
γ: discount rate (0<γ≦1)

Then, in selecting the action A _t in (B) above, a value function created by past learning is used to select the action A _t that optimizes the future reward (R _t+1 +R _t+2 + ...) in the current state S _t , that is, the action A _t with the highest value in state S _t . In Q-learning, the action A _t that optimizes the reward (R _t+1 +R _t+2 + ...) corresponds to the action A _t that maximizes the reward (R _t+1 +R _t+2 + ...).

Various reinforcement learning algorithms are well known, including Q-learning, SARSA, TD learning, and AC, and any of these reinforcement learning algorithms may be adopted as the method applied to the present invention. Because these reinforcement learning algorithms are well known, further detailed explanation of each algorithm will be omitted in this specification.

Furthermore, as a method for updating the value function, it is not necessary to adopt a learning method in which the value function is updated each time a certain action A _t is applied to the current state S _t , thereby transitioning the state S _t to a new state S t ₊₁ , i.e., so-called online learning. For example, it is possible to adopt so-called batch learning or mini-batch learning, in which the process of applying a certain action A _t to the current state S _t to transition the state S _t to a new state S _t+1 is repeated, these states and actions are accumulated as learning data, and the accumulated learning data is used to update the value function.

For this purpose, the controller 22 includes an injection amount output unit 221, a state value acquisition unit 222, a reward calculation unit 223, and a learning unit 224.

The injection quantity output unit 221 outputs the injection quantity as a command value for the cleaning device 21. The injection quantity becomes the operating value described below. The state value acquisition unit 222 acquires the surface temperature of the heat transfer surface obtained from images captured by the infrared cameras 11 and 12 and the gas temperature Tg measured by the thermometer 13 as state values. The acquired state values are stored in the storage of the controller 22. The storage stores state values for each specified time period as history. Furthermore, the state value acquisition unit 222 also acquires an injection quantity command value and stores it in the storage. The storage stores the position of the heat transfer surface cleaned by the cleaning device 21, the time of cleaning, and the amount of water sprayed as history. As a result, the remuneration calculation unit 223 regards the surface temperature Td and gas temperature Tg of the heat transfer surface at a specified heat transfer surface position included in the period from the time cleaning was performed at that heat transfer surface position until a specified time before as the pre-cleaning surface temperature and pre-cleaning gas temperature, and regards the surface temperature Td and gas temperature Tg of the heat transfer surface at that heat transfer surface position included in the period from the time cleaning was performed until a specified time after as the post-cleaning surface temperature and post-cleaning gas temperature, and performs the following calculations.

The reward calculation unit 223 analyzes the state values acquired by the state value acquisition unit 222 based on preset reward conditions and calculates the reward. The reward calculation unit 223 outputs the calculated reward to the learning unit 224.

Reward conditions are conditions for providing a reward in reinforcement learning. The reward conditions are stored in advance in storage. For example, reward conditions include providing a larger reward when the post-cleaning gas temperature is determined to be low than when the post-cleaning gas temperature is determined to be high. Also, for example, reward conditions include providing a larger reward when the post-cleaning surface temperature is determined to be high than when the post-cleaning surface temperature is determined to be low. Rewards can be, for example, a positive reward, a negative reward, or no reward, i.e., zero reward.

The reward calculation unit 223 calculates a positive reward if the post-cleaning gas temperature is less than a predetermined first reference value, and calculates a negative or zero reward if it is greater than the first reference value. Furthermore, the reward calculation unit 223 calculates a positive reward if the post-cleaning surface temperature is less than a predetermined second reference value, and calculates a negative or zero reward if it is greater than the second reference value. The first and second reference values are determined based on simulations, past operating results, etc., and are pre-stored in the controller 22's storage.

Instead of comparing the post-cleaning gas temperatures themselves, the reward calculation unit 223 may compare the value obtained by subtracting the post-cleaning gas temperature from the pre-cleaning gas temperature with a predetermined third reference value, and calculate a positive reward if the value is greater than the third reference value, and a negative or zero reward if the value is less than the third reference value. Similarly, instead of comparing the post-cleaning surface temperatures themselves, the reward calculation unit 223 may compare the value obtained by subtracting the post-cleaning surface temperature from the pre-cleaning surface temperature with a predetermined fourth reference value, and calculate a positive reward if the value is greater than the fourth reference value, and a negative or zero reward if the value is less than the fourth reference value.

The reward calculation does not have to be performed for cleaning at all heat transfer surface positions. For example, if the cleaning device 21 cleans a heat transfer surface position that has been predetermined as the target of calculation because the dust thickness estimated by the calculator 14 exceeds a predetermined threshold value, the reward calculation unit 223 may calculate the reward from the amount of water sprayed, the post-cleaning surface temperature, and the post-cleaning gas temperature related to that cleaning.

The learning unit 224 performs machine learning to determine an operating value, i.e., the injection amount to be output by the injection amount output unit 221, from the current state value based on the injection amount command value to the cleaning device 21, the state value acquired from the state value acquisition unit 222, and the reward calculated by the reward calculation unit 223. More specifically, the learning unit 224 updates the value function, for example, a state-action value function Q(S _t , A _t ), in which the state S _t defined by a combination of state values and the determination of the operating value in that state are expressed as arguments, so as to maximize the reward obtained.

The results of learning by the learning unit 224 are stored in storage as a learned model. Note that methods of storing a value function as a learning result generally include a method using an approximation function and a method using an array. However, in addition to these methods, for example, when the state takes on many states, a method using a supervised learning device such as a multi-value output SVM or neural network that inputs the state S _t and the action A _{t and} outputs a value may also be used.

The injection quantity output unit 221 outputs the injection quantity, which is an operating value, based on the trained model and the current state value.

The trained model may be a common trained model that is independent of the position of the heat transfer surface, or may include multiple trained models that differ depending on the position or area of the heat transfer surface. For example, the trained models used to determine the injection amount may be different for the radiation chamber 119, the first flue 120, and the second flue 121. In this case, the learning unit 224 performs machine learning based on the state value in the radiation chamber 119 and the injection amount within the radiation chamber 119. The same applies to the first flue 120 and the second flue 121. Furthermore, for example, the trained models used to determine the injection amount may be different for the first heat transfer surface F1, the second heat transfer surface F2, the third heat transfer surface F3, and the fourth heat transfer surface F4.

Next, we will explain the flow of machine learning in this embodiment. Figure 6 is a flowchart showing the flow of machine learning performed by the controller shown in Figure 5.

When machine learning starts, in step S1, the state value acquisition unit 222 acquires a state value at time t as data for identifying a state S _t . The state value acquisition unit 222 stores the acquired state value in storage. In step S2, the learning unit 224 identifies a current state S _t based on the state value acquired by the state value acquisition unit 222. In step S3, the learning unit 224 selects an action A _t based on past learning results and the identified state S _t . The action A _t is a determination of an operating value according to the defined state S _t . Here, for example, multiple operating values may be prepared as selectable actions, and the action A that will maximize the reward R to be obtained in the future may be selected based on past learning results. The injection amount output unit 221 determines an operating value to be output based on the selected action A _t and outputs the determined operating value to the cleaning device 21. In step S4, the cleaning device 21 executes the action A _t , and the interior of the boiler 104 is cleaned.

After the action A _t is executed and the state transitions, the state value acquisition unit 222 acquires data for identifying the state S _t+1 as a state value in step S5. At this stage, the state of the boiler 104 changes over time from time t to time t+1 due to the executed action A _t . In step S6, the reward calculation unit 223 calculates a reward R _t+1 from the state value at time t+1 based on the set reward condition. In step S7, the learning unit 224 performs machine learning based on the state S _t , the action A _t , and the reward R _t+1 . The value function used for learning is determined according to the learning algorithm to be applied. In step S8, the learning unit 224 stores the learning results in storage. Learning at time t+1 can be performed in the same manner as described above, with time t in the flowchart shown in FIG. 6 replaced with time t+1.

In this way, the learning unit 224 repeats machine learning. Learning by the learning unit 224 may be completed when it is confirmed that an optimal system for injection amount control has been achieved. For example, learning by the learning unit 224 may be completed when the reward calculated by the reward calculation unit 223 becomes positive by a predetermined percentage or more. After learning is complete, no further learning is performed using newly obtained state values and operating values, and the learned model is not updated.

On the other hand, machine learning in the learning unit 224 may continue without being completed. The optimal operating values may change over time due to changes in the cleaning device 21 and boiler 104. By continuing machine learning, it is possible to respond to changes in the optimal operating values.

As described above, with the heat transfer surface cleaning system 2 of this embodiment, the surface temperatures of the heat transfer surfaces F1, F2, F3, and F4 obtained from images captured by the infrared cameras 11 and 12 and the gas temperature Tg inside the boiler 104 are used as state values, and a reward is calculated based on the gas temperature Tg after cleaning by the cleaning device 21, thereby learning the optimal amount of water sprayed from the cleaning device 21, which is an operating value. This allows the dust thickness td estimated from the heat transfer surface surface temperature Td and gas temperature Tg to be pinpoint correlated with the amount of water sprayed to clean the dust for each position on the heat transfer surface according to the surface temperature distribution of the heat transfer surface. Therefore, a cleaning capacity appropriate to the thickness of dust adhering to the heat transfer surface can be applied.

Furthermore, in this embodiment, as described above, the reward calculation unit 223 calculates a positive reward if the post-cleaning gas temperature is less than a predetermined first reference value, and calculates a negative or zero reward if it is greater than the first reference value. Alternatively, the reward calculation unit 223 calculates a positive reward if the value obtained by subtracting the pre-cleaning gas temperature from the post-cleaning gas temperature is less than a predetermined third reference value, and calculates a negative or zero reward if it is greater than the third reference value. In this way, the decrease in gas temperature Tg due to cleaning can be evaluated as a decrease in dust thickness td and an increase in thermal efficiency between the exhaust gas in the boiler 104 and the heat transfer tube 123.

Furthermore, in this embodiment, in addition to the post-cleaning gas temperature, the reward is also calculated based on the post-cleaning surface temperature. This allows for more appropriate learning using multiple evaluation criteria. In this case, the reward calculation unit 223 calculates a positive reward if the post-cleaning surface temperature is greater than a predetermined second reference value, and a negative or zero reward if it is less than the second reference value. Alternatively, the reward calculation unit 223 calculates a positive reward if the value obtained by subtracting the pre-cleaning surface temperature from the post-cleaning surface temperature is greater than a predetermined fourth reference value, and a negative or zero reward if it is less than the fourth reference value. This allows for evaluation of the increase in the surface temperature of the heat transfer surface due to cleaning as a reduction in the dust thickness td and an increase in the thermal efficiency between the exhaust gas in the boiler 104 and the heat transfer tube 123.

Furthermore, in this embodiment, a water jet cleaning device that sprays water is used as the cleaning device 21. This makes it possible to clean small areas, allowing for more pinpoint cleaning.

[Modification]
Although the embodiments of the present disclosure have been described above, the present disclosure is not limited to the above-described embodiments, and various improvements, changes, and modifications are possible within the scope of the spirit of the present disclosure.

For example, while the above embodiment shows an example of application to a waste heat boiler that uses exhaust gas from an incineration plant 100, the configuration of the present disclosure can be widely applied to solid fuel-fired boilers that generate dust.

In addition, while the above embodiment illustrates an example in which infrared cameras 11 and 12 are installed on the first heat transfer surface F1 and the second heat transfer surface F2, an infrared camera may also be installed on the third heat transfer surface F3 or the fourth heat transfer surface F4 in addition to or instead of this. Furthermore, at least one infrared camera may be installed to obtain the surface temperature distribution on at least one heat transfer surface.

In addition, in the above embodiment, an example was shown in which the computing unit 14 of the heat transfer surface monitoring device 1 and the controller 22 of the heat transfer surface cleaning system 2 are configured as separate computers, but this is not limited to this. For example, the controller 22 of the heat transfer surface cleaning system 2 may have the computing unit 14 of the heat transfer surface monitoring device 1. Also, for example, the control device of the incineration plant 100 may have the computing unit 14 and the controller 22.

Furthermore, in the above embodiment, an example was given in which the controller 22 has functional blocks or circuits 222, 223, and 224 for performing machine learning, but the heat transfer surface cleaning system 2 may also be provided with a separate learning device for performing machine learning.

Furthermore, in the above embodiment, an example was given in which the remuneration calculation unit 223 calculates the remuneration based on the post-cleaning surface temperature in addition to the post-cleaning gas temperature, but the remuneration calculation unit 223 may also calculate the remuneration based only on the post-cleaning gas temperature.

In addition, in the above embodiment, the cleaning device 21 is a water jet cleaning device equipped with a water injection nozzle 21a that injects water as a cleaning fluid, but this is not limited to this. For example, the cleaning device 21 may be a steam cleaning device equipped with a steam injection nozzle that injects steam as a cleaning fluid. Furthermore, for example, the cleaning device 21 may be a pressure wave cleaning device equipped with an exhaust duct that injects pressure waves as a cleaning fluid. Multiple types of cleaning devices may be combined. For example, the cleaning device 21 that cleans the radiation chamber 119 and the first flue 120 may be a water jet cleaning device, and the cleaning device 21 that cleans the second flue 121, into which the superheater protrudes, may be a pressure wave cleaning device or a steam cleaning device.

In addition, while the above embodiment illustrates a configuration in which the cleaning device 21 supplies cleaning fluid into the boiler 104 through the ceiling of the boiler 104, this is not limiting. For example, the cleaning device 21 may supply cleaning fluid into the boiler 104 through the side wall of the boiler 104. For example, at least one of the heat transfer surfaces F1, F2, F3, and F4 may have an insertion hole through which the water supply pipe 21c is inserted. In this case, the heat transfer surface may have multiple insertion holes to enable cleaning of multiple heat transfer surface areas.

In addition, while the above embodiment illustrates a configuration in which the incineration plant 100 is equipped with both the heat transfer surface monitoring device 1 and the heat transfer surface cleaning system 2, the heat transfer surface cleaning system 2 can also be applied to boilers in incineration plants and the like that do not have the heat transfer surface monitoring system 1. In other words, the heat transfer surface cleaning system 2 may be applied to cleaning the heat transfer surfaces in the boiler 104 without estimating the thickness of dust adhering to the heat transfer surface from the surface temperature distribution of the heat transfer surface and the gas temperature. For example, the cleaning device 21 may perform cleaning at a predetermined timing. The trained model described above may be used to control the amount of water sprayed at this time, and machine learning may be performed using that amount of sprayed water.

Summary of the Disclosure
A heat transfer surface cleaning system (2) according to one aspect of the present disclosure includes infrared cameras (11, 12) that measure the surface temperatures of heat transfer surfaces (F1, F2, F3, F4) exposed inside a boiler (104) by photographing the heat transfer surfaces (F1, F2, F3, F4), a thermometer (13) that measures the gas temperature in the internal space of the boiler (104), a cleaning device (21) that cleans the heat transfer surfaces (F1, F2, F3, F4) by spraying a cleaning fluid, and a controller (22) that controls the spray amount of the cleaning fluid. ), wherein the controller (22) outputs the injection amount, acquires a state value including the measured surface temperature of the heat transfer surface (F1, F2, F3, F4) and the gas temperature, acquires a post-cleaning gas temperature measured by the thermometer (13) after cleaning by the cleaning device (21) based on the injection amount, calculates a reward based on the post-cleaning gas temperature, and performs machine learning to determine the injection amount to be output for the current state value based on the injection amount, the state amount, and the reward.

With the above configuration, the surface temperatures of the heat transfer surfaces F1, F2, F3, and F4 obtained from images captured by the infrared cameras 11 and 12 and the gas temperature Tg inside the boiler 104 are used as state values, and the reward is calculated based on the gas temperature Tg after cleaning by the cleaning device 21, thereby learning the optimal amount of water sprayed from the cleaning device 21, which is an operating value. This allows the dust thickness td estimated from the heat transfer surface surface temperature Td and gas temperature Tg to be pinpointedly correlated with the amount of water sprayed to clean the dust for each position on the heat transfer surface according to the surface temperature distribution of the heat transfer surface. Therefore, it is possible to apply a cleaning capacity appropriate to the thickness of dust adhering to the heat transfer surface.

The controller (22) may calculate a positive reward if the post-cleaning gas temperature is less than a predetermined first reference value, and may calculate a negative or zero reward if the post-cleaning gas temperature is greater than the first reference value. This allows the reduction in gas temperature Tg due to cleaning to be evaluated as a reduction in dust thickness and an increase in thermal efficiency between the exhaust gas in the boiler 104 and the heat transfer tube 123.

In addition to the post-cleaning gas temperature, the controller (22) may also acquire the post-cleaning surface temperature of the heat transfer surfaces (F1, F2, F3, F4) measured by the infrared camera (11, 12) after cleaning by the cleaning device (21) based on the injection amount, and calculate the reward based on the post-cleaning gas temperature and the post-cleaning surface temperature. This allows for more appropriate learning using multiple evaluation criteria.

The controller (22) may calculate a positive reward if the post-cleaning surface temperature is less than a predetermined second reference value, and may calculate a negative or zero reward if the post-cleaning surface temperature is greater than the second reference value. This allows the reduction in the surface temperature of the heat transfer surface due to cleaning to be evaluated as a reduction in the thickness of dust and an increase in heat exchange between the exhaust gas in the boiler 104 and the heat transfer tube 123.

The controller (22) may calculate a positive reward if the value obtained by subtracting the gas temperature after cleaning from the gas temperature before cleaning is greater than a predetermined third reference value, and may calculate a negative or zero reward if the value is less than the third reference value. This allows the decrease in gas temperature Tg due to cleaning to be evaluated as a reduction in dust thickness and an increase in heat exchange between the exhaust gas in the boiler 104 and the heat transfer tubes 123.

The reward calculation unit (223) may calculate a positive reward if the value obtained by subtracting the surface temperature after cleaning from the surface temperature of the heat transfer surfaces (F1, F2, F3, F4) before cleaning is greater than a predetermined fourth reference value, and may calculate a negative or zero reward if the value is less than the fourth reference value. This allows the reduction in the surface temperature of the heat transfer surfaces due to cleaning to be evaluated as a reduction in the thickness of dust and an increase in heat exchange between the exhaust gas in the boiler 104 and the heat transfer tubes 123.

The cleaning device (21) may be equipped with a water injection nozzle (21a) that injects water as the cleaning fluid.

2 Heat transfer surface cleaning system 11, 12 Infrared camera 13 Thermometer 21 Cleaning device 21a Water injection nozzle 22 Controller 104 Boiler F1, F2, F3, F4 Heat transfer surface

Claims (7)

an infrared camera that measures the surface temperature of a heat transfer surface exposed inside the boiler by photographing the heat transfer surface;
a thermometer for measuring a gas temperature in the internal space of the boiler;
a cleaning device that cleans the heat transfer surface by spraying a cleaning fluid;
a controller that controls the amount of spray of the cleaning fluid,
The controller
outputting the injection amount;
acquiring state values including the measured surface temperature distribution of the heat transfer surface and the gas temperature;
After cleaning by the cleaning device based on the injection amount, a post-cleaning gas temperature measured by the thermometer is acquired, and a reward is calculated based on the post-cleaning gas temperature.
A heat transfer surface cleaning system that performs machine learning to determine the injection amount to be output at the current state value based on the injection amount, the state value , and the reward. The heat transfer surface cleaning system of claim 1, wherein the controller calculates a positive reward if the post-cleaning gas temperature is less than a predetermined first reference value, and calculates a negative or zero reward if the post-cleaning gas temperature is greater than the first reference value. The heat transfer surface cleaning system of claim 1 or 2, wherein the controller acquires, in addition to the post-cleaning gas temperature, the post-cleaning surface temperature of the heat transfer surface measured by the infrared camera after cleaning by the cleaning device based on the injection amount, and calculates the reward based on the post-cleaning gas temperature and the post-cleaning surface temperature. The heat transfer surface cleaning system of claim 3, wherein the controller calculates a positive reward if the post-cleaning surface temperature is less than a predetermined second reference value, and calculates a negative or zero reward if the post-cleaning surface temperature is greater than the second reference value. The heat transfer surface cleaning system of claim 1, wherein the controller calculates a positive reward if the value obtained by subtracting the gas temperature after cleaning from the gas temperature before cleaning is greater than a predetermined third reference value, and calculates a negative or zero reward if the value is less than the third reference value. The heat transfer surface cleaning system of claim 3, wherein the controller calculates a positive reward if the value obtained by subtracting the surface temperature after cleaning from the surface temperature of the heat transfer surface before cleaning is greater than a predetermined fourth reference value, and calculates a negative or zero reward if the value is less than the fourth reference value. A heat transfer surface cleaning system according to any one of claims 1 to 6, wherein the cleaning device is equipped with a water injection nozzle that injects water as the cleaning fluid.