System and method for removing volatile organic compounds

Abstract

The invention relates to a system and a method for removing a volatile organic compound. The method comprises the following steps of: removing the volatile organic compound; the VOC removal system removes VOC from the waste liquid of the semiconductor manufacturing process. The VOC removal system measures current VOC removal parameters and transmits the current VOC removal parameters to the analysis model trained in the machine learning process. The analysis model predicts future VOC removal efficiency based on the current VOC removal parameters. The analysis model generates an adjustment parameter based on the current VOC removal parameter and the predicted future VOC removal efficiency. The control system adjusts the VOC removal system based on the adjustment parameters. The VOC removal system can effectively remove VOC from the process gas.

Description

[0001] This application is a divisional application of the patent application with the application date of February 25, 2022, application number 202210179916.2, and invention title "Volatile Organic Compound Removal System and Method". Technical Field

[0002] This disclosure relates to the field of purification of waste liquid in semiconductor manufacturing processes. Background Art

[0003] Multiple semiconductor manufacturing processes are used to fabricate integrated circuits. Such processes may include thin film deposition processes, etching processes, annealing processes, lithography processes, and many other types of processes. Semiconductor manufacturing processes often generate waste liquid. These waste liquids sometimes include high-concentration volatile organic compounds (VOCs). Volatile organic compounds are harmful if released into the atmosphere. Summary of the Invention

[0004] According to some embodiments of the present disclosure, a method for removing volatile organic compounds (VOCs) includes: measuring the VOC concentration of the exhaust gas of a semiconductor manufacturing process using a VOC concentration sensor; after measuring the VOC concentration of the exhaust gas of the semiconductor manufacturing process, measuring the temperature of the exhaust gas of the semiconductor manufacturing process passing through the VOC concentration sensor using a temperature sensor; after measuring the VOC concentration of the exhaust gas of the semiconductor manufacturing process and measuring the temperature of the exhaust gas of the semiconductor manufacturing process, removing a plurality of VOCs from the exhaust gas of the semiconductor manufacturing process using a VOC removal system, wherein the temperature sensor is located between the VOC concentration sensor and the VOC concentrator rotor of the VOC removal system; providing a plurality of parameters of the VOC removal system to an analysis model trained by a machine learning process, the parameters including the temperature and VOC concentration of the exhaust gas; generating a predicted future VOC removal efficiency based on the parameters using the analysis model; generating a plurality of adjustment parameters based on the predicted future VOC removal efficiency and the parameters using the analysis model, wherein the adjustment parameters include the recommended temperature of the VOC burner; adjusting the VOC removal system based on the adjustment parameters; and directly receiving a first portion of the exhaust gas from the VOC concentrator rotor and a second portion of the exhaust gas from the heat exchanger of the VOC removal system through the exhaust chimney of the VOC removal system.

[0005] In some embodiments according to the present disclosure, a volatile organic compound (VOC) removal system includes a VOC concentrator rotor, a first temperature sensor, a second temperature sensor, a desorption fan, a burner, a control system, a heat exchanger, and an exhaust chimney. The VOC concentrator rotor has an adsorption zone, a desorption zone, and a cooling zone. The cooling zone connects the adsorption zone to the desorption zone. The adsorption zone is configured to adsorb a plurality of VOCs from a first portion of the exhaust gas from a semiconductor process, and the desorption zone is configured to desorb the VOCs to a second portion of the exhaust gas. The first temperature sensor is placed on the output side of the adsorption zone of the VOC concentrator rotor and is configured to sense a first temperature of the VOC concentrator rotor. The second temperature sensor is placed on the input side of the adsorption zone of the VOC concentrator rotor and is configured to sense a second temperature of the VOC concentrator rotor. Each of the first temperature sensor and the second temperature sensor is closer to the adsorption zone than the desorption zone. The desorption fan is configured to drive the second portion of the exhaust gas through the desorption zone of the VOC concentrator rotor. The burner is configured to burn the VOCs from the second portion of the exhaust gas. The control system includes an analysis model. The analysis model is trained through a machine learning process and is configured to analyze a plurality of parameters of the VOC concentrator rotor, the desorption fan, and the burner, and is configured to generate a predicted future VOC removal efficiency based on the parameters, and generate a plurality of adjustment parameters based on the parameters and the predicted future VOC removal efficiency. The control system is configured to adjust the operations of the VOC concentrator rotor, the desorption fan, and the burner based on the adjustment parameters. The parameters include the temperature of the VOC concentrator rotor, and the adjustment parameters include the recommended temperature of the burner. The parameters include the first temperature sensed by the first temperature sensor and the second temperature sensed by the second temperature sensor. The heat exchanger is configured to receive heat from the second portion of the exhaust gas downstream of the burner. The exhaust chimney is configured to directly receive the first portion of the exhaust gas from the VOC concentrator rotor and directly receive the second portion of the exhaust gas from the heat exchanger.

[0006] In some embodiments according to the present disclosure, a method for removing volatile organic compounds (VOCs) includes: measuring a first temperature of a VOC concentrator rotor using a first temperature sensor placed on the input side of the adsorption zone of the VOC concentrator rotor; transferring an exhaust gas from a semiconductor process to the VOC concentrator rotor; adsorbing a plurality of VOCs from a first portion of the exhaust gas using the VOC concentrator rotor; transferring a second portion of the exhaust gas from the VOC concentrator rotor to a heater; transferring the second portion of the exhaust gas from the heater to the VOC concentrator rotor; desorbing VOCs from the VOC concentrator rotor into the second portion of the exhaust gas by operating a desorption fan adjacent to the VOC concentrator rotor; combusting the VOCs from the second portion of the exhaust gas using a burner; measuring a second temperature of the VOC concentrator rotor using a second temperature sensor placed on the output side of the adsorption zone of the VOC concentrator rotor, wherein a portion of the adsorption zone of the VOC concentrator rotor is located between the first temperature sensor and the second temperature sensor; generating a predicted future VOC removal efficiency at least partially based on the first temperature and the second temperature of the VOC concentrator rotor using an analysis model trained through a machine learning process; generating a plurality of adjustment parameters based on the predicted future VOC removal efficiency and the first temperature and the second temperature of the VOC concentrator rotor using the analysis model, wherein the adjustment parameters include a recommended temperature of the burner; adjusting the rotational speed of the VOC concentrator rotor and the temperature of the burner based on the adjustment parameters; and receiving the first portion of the exhaust gas directly from the VOC concentrator rotor and the second portion of the exhaust gas directly from a heat exchanger using an exhaust chimney of the VOC removal system. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 is a block diagram of a volatile organic compound (VOC) removal system according to an embodiment;

[0008] Figure 2 is a block diagram of a control system of a VOC removal system according to an embodiment;

[0009] Figure 3 is a block diagram of an analysis model according to an embodiment;

[0010] Figure 4 is a flowchart of a process for training an encoder of an analysis model through a machine learning process;

[0011] Figure 5 is a flowchart of a process for training a decoder of an analysis model through a machine learning process;

[0012] Figure 6 is a flowchart of a method for operating a VOC removal system according to an embodiment;

[0013] Figure 7 is a flowchart of a method for operating a VOC removal system according to an embodiment.

[0014]

Symbol Explanation

[0015] 100: VOC removal system

[0016] 102: Semiconductor process

[0017] 104: VOC concentrator rotor

[0018] 106: Adsorption zone

[0019] 108: Cooling zone

[0020] 110: Desorption zone

[0021] 112: Heat exchanger

[0022] 114: Exhaust chimney

[0023] 116: Desorption fan

[0024] 118: Heat exchanger

[0025] 120: Burner

[0026] 122, 124: Temperature sensors

[0027] 126: Control system

[0028] 128: Waste liquid

[0029] 130: First part of the waste liquid

[0030] 132: Second part of the waste liquid

[0031] 134, 136: VOC sensors

[0032] 140: Analysis model

[0033] 141: Training module

[0034] 142: Training set data

[0035] 144: Historical VOC removal efficiency data

[0036] 146: Historical VOC removal condition data

[0037] 148: Processing resources

[0038] 150: Memory resources

[0039] 152: Communication resources

[0040] 153: Data integration server

[0041] 160: Encoder

[0042] 162: Decoder

[0043] 164: Process condition vector

[0044] 166: Data fields of the process condition vector

[0045] 168a to 168e: Neural layers

[0046] 168f: Final neural layer

[0047] 170: Node

[0048] 172: Predicted VOC removal efficiency

[0049] 174: Predicted process condition vector

[0050] 175: Data fields of the predicted process condition vector

[0051] 400, 500: Process

[0052] 402, 404, 406, 408, 410, 412, 414, 505, 504, 506, 508, 510, 512, 514, 602, 604, 606, 608, 610, 702, 704, 706, 708, 710, 712, 714, 716, 718, 720: Steps 600, 700: Method Detailed implementation manners

[0053] In the following description, many thicknesses and materials of various layers and structures within an integrated circuit die are described. Specific dimensions and materials of various embodiments are given as examples. According to this disclosure, those skilled in the art will recognize that in many cases other dimensions and materials can be used without departing from the scope of this disclosure.

[0054] The following disclosure provides many different embodiments or examples for implementing different features of the described subject matter. Specific examples of components and configurations are described below to simplify this description. Of course, these are only examples and are not intended to be restrictive. For example, forming the first feature on or above the second feature in the following description may include embodiments in which the first feature and the second feature are formed in direct contact, and may also include embodiments in which additional features may be formed between the first feature and the second feature such that the first feature and the second feature may not be in direct contact. Additionally, this disclosure may repeat reference numerals and / or letters in various examples. This repetition is for simplicity and clarity purposes and does not itself indicate a relationship between the various embodiments and / or configurations discussed.

[0055] In addition, for ease of description, spatially relative terms such as "below", "beneath", "lower", "above", "upper", etc. may be used herein to describe the relationship of one element or feature to another (or other) element or feature, as illustrated in the figures. In addition to the orientation depicted in the figures, the spatially relative terms are also intended to encompass different orientations of the device during use or operation. The device may be oriented in other ways (rotated 90 degrees or in other orientations), and likewise the spatially relative descriptors used herein may be interpreted accordingly.

[0056] In the following description, certain specific details are set forth in order to provide a thorough understanding of the various embodiments of the present disclosure. However, those skilled in the art will understand that the present disclosure may be practiced without these specific details. In other instances, well-known structures associated with electronic components and manufacturing techniques have not been described in detail so as not to unnecessarily obscure the description of the embodiments of the present disclosure.

[0057] Unless the context requires otherwise, throughout the following specification and claims, the word "comprise" and its variations (such as "comprises" and "comprising") shall be construed in an open, inclusive sense, i.e., to mean "including but not limited to".

[0058] The use of ordinal numbers such as first, second, and third does not necessarily imply a sense of ordering, but may only distinguish multiple instances of an action or structure.

[0059] References to "an embodiment" or "one embodiment" in the present specification throughout mean that the particular features, structures, or characteristics described in connection with that embodiment are included in at least one embodiment. Thus, the appearances of the phrases "in an embodiment" or "in one embodiment" throughout the present specification are not necessarily all referring to the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0060] As used in this specification and the appended claims, the singular forms "a" and "the" include plural referents unless the context clearly dictates otherwise. It should also be noted that the term "or" is generally used in its "and / or" sense unless the context clearly dictates otherwise.

[0061] Embodiments of the present disclosure provide a VOC removal system for effectively removing volatile organic compounds (VOCs) from semiconductor process waste liquids. Embodiments of the present disclosure utilize a machine learning process to periodically adjust the operation of the VOC removal system to maintain a high VOC removal efficiency. In particular, the machine learning model receives sensor data related to the current operating parameters of the VOC removal system and data related to the current VOC removal efficiency and identifies adjustments to be made to the operating parameters to improve or maintain the VOC removal efficiency. This provides many benefits to the VOC removal system. In particular, the efficiency of removing hazardous VOCs from the waste liquid is much higher than in traditional VOC removal systems. As a result, the emissions from semiconductor processing facilities are cleaner. This leads to improved environmental conditions.

[0062] Figure 1 FIG. 4 is a block diagram of a VOC removal system 100 according to an embodiment. The VOC removal system 100 includes a concentrator rotor 104. The concentrator rotor 104 includes an adsorption zone 106, a cooling zone 108, and a desorption zone 110. The VOC removal system 100 further includes a heat exchanger 112, an exhaust stack 114, a desorption fan 116, a heat exchanger 118, a burner 120, temperature sensors 122 and 124, and a control system 126. As will be described in more detail below, the control system 126 utilizes a machine learning process to periodically adjust the operation of the VOC removal system 100 to maintain or improve the VOC removal efficiency.

[0063] A semiconductor process 102 generates a waste liquid 128. Examples of the semiconductor process 102 may include a thin film deposition process, an etching process, a doping process, an annealing process, or other thin film deposition processes. In addition, the scope of the present disclosure is not limited to waste liquids generated by semiconductor processes. Other processes (such as thin film transistor processes or other processes for forming thin film electronic devices such as LCD screen devices and other types of devices) may result in a waste liquid 128 containing VOCs. The waste liquid 128 may contain a high concentration of VOCs. It is beneficial to remove VOCs from the waste liquid 128 before releasing the waste liquid into the atmosphere. The VOC removal system 100 removes VOCs from the waste liquid 128.

[0064] The waste liquid 128 is transferred to the VOC concentrator rotor 104. In one embodiment, the VOC concentrator rotor 104 has the shape of a large wheel with a diameter between 3 m and 5 m, but other sizes may be utilized without departing from the scope of the present disclosure.

[0065] The VOC concentrator rotor 104 includes an adsorption zone 106, a cooling zone 108, and a desorption zone 110. More appropriately, it can be stated that the VOC concentrator rotor 104 rotates through the adsorption zone 106, the cooling zone 108, and the desorption zone 110. This is because the VOC concentrator rotor 104 rotates slowly. The adsorption zone 106, the cooling zone 108, and the desorption zone 110 remain stationary. The VOC concentrator rotor 104 rotates through each of these zones. In particular, the VOC concentrator rotor 104 rotates from the adsorption zone 106 to the desorption zone 110, from the desorption zone 110 to the cooling zone 108, and from the cooling zone 108 back to the adsorption zone 106.

[0066] In one embodiment, the VOC concentrator rotor 104 includes a calcined ceramic honeycomb substrate. The honeycomb substrate may include an inorganic binder. Hydrated aluminum silicate (also known as zeolite) is embedded in the ceramic honeycomb substrate. The VOC concentrator rotor 104 having this structure and material is capable of adsorbing VOCs from the exhaust gas at low temperatures and desorbing VOCs into the exhaust gas at high temperatures. Other materials and combinations of materials may be used for the VOC concentrator rotor 104 without departing from the scope of this disclosure.

[0067] When the waste liquid 128 is transferred from the semiconductor process 102 to the VOC concentrator rotor 104, a first portion 130 of the waste liquid 128 enters the adsorption zone 106. A second portion 132 of the waste liquid 128 enters the cooling zone 108 of the VOC concentrator rotor 104.

[0068] When the first portion 130 of the waste liquid 128 passes through the adsorption zone 106, VOCs are adsorbed onto the VOC concentrator rotor 104. The first portion 130 of the waste liquid 128 contains a high concentration of VOCs before passing through the adsorption zone 106 of the VOC concentrator rotor 104. After the first portion 130 of the waste liquid 128 is transferred from the adsorption zone 106 of the VOC concentrator rotor 104, the VOCs have been removed from the first portion 130 of the waste liquid 128. In particular, the VOCs from the first portion 130 of the waste liquid 128 are adsorbed onto that portion of the concentrator rotor 104 passing through the adsorption zone 106.

[0069] The first portion 130 of the waste liquid 128 is transferred from the VOC concentrator rotor 104 to the exhaust chimney 114. The exhaust chimney 114 transfers the first portion 130 of the waste liquid 128 to the atmosphere. Since the VOC concentrator rotor 104 has adsorbed VOCs from the first portion 130 of the waste liquid 128, the first portion of the waste liquid 128 is purified when it enters the atmosphere via the exhaust chimney 114.

[0070] As the second portion 132 of the waste liquid 128 passes through the cooling zone 108, very little VOC is adsorbed from the second portion 132 of the waste liquid 128 onto the VOC concentrator rotor 104. This is because that portion of the VOC concentrator rotor 104 is relatively hot, the reason for which will be described further below. Because the VOC concentrator rotor 104 is at a high temperature, that portion of the VOC concentrator rotor 104 does not adsorb a great deal of VOC from the second portion 132 of the waste liquid 128. As it passes through the cooling zone 108, the temperature of that portion of the VOC concentrator rotor 104 passing through the cooling zone 108 gradually decreases. That portion of the VOC concentrator rotor 104 that is transferred from the cooling zone 108 to the adsorption zone 106 is cold enough to adsorb VOC from the first portion of the waste liquid 128 in the adsorption zone 106.

[0071] The second portion 132 of the waste liquid 128 is transferred from the cooling zone 108 to the heat exchanger 112. The heat exchanger 112 heats the second portion 132 of the waste liquid 128 to a temperature between 190 °C and 230 °C. As will be described in more detail below, the heated second portion 132 of the waste liquid 128 will enable the VOC concentrator rotor 104 to desorb VOC into the heated second portion 132 of the waste liquid 128. The heat exchanger 112 may heat the second portion 132 of the waste liquid 128 to a temperature other than those described above without departing from the scope of this disclosure.

[0072] In one embodiment, the heat exchanger 112 is replaced by a different type of heater. Any suitable heater can be used to heat the second portion 132 of the waste liquid 128 to a temperature at which the VOC concentrator rotor 104 can desorb VOC into the second portion 132 of the waste liquid 128.

[0073] The second portion 132 of the waste liquid 128 is transferred from the heat exchanger 112 to the desorption zone 110 of the VOC concentrator rotor 104. Because the second portion 132 of the waste liquid 128 is at a high temperature as described above, the VOC concentrator rotor 104 desorbs VOC from the VOC concentrator rotor 104 into the second portion 132 of the waste liquid 128. Thus, the VOC adsorbed from the first portion 130 of the waste liquid 128 is desorbed onto the second portion 132 of the waste liquid 128.

[0074] Rotation of the VOC concentrator rotor 104 enables continuous removal of VOCs from the waste liquid 128 without replacing the VOC concentrator rotor 104. After adsorbing VOCs from the first portion 130 of the waste liquid 128 at the adsorption zone 106, rotation of the VOC concentrator rotor 104 causes the VOC-containing portion of the VOC concentrator rotor 104 to rotate to the desorption zone 110. At the desorption zone 110, the VOC-containing portion of the VOC concentrator rotor 104 desorbs the VOCs into the second portion 132 of the waste liquid 128. The VOC concentrator rotor 104 is heated by the heated second portion 132 of the waste liquid 128 at the desorption zone 110. The portion of the VOC concentrator rotor 104 leaving the desorption zone 110 no longer contains VOCs because the VOCs have been desorbed onto the second portion 132 of the waste liquid 128. The heated portion of the VOC concentrator rotor 104 cools in the cooling zone 108 and returns to the adsorption zone 106, ready to adsorb more VOCs from the first portion 130 of the waste liquid 128. In this way, the VOC concentrator rotor 104 can operate continuously.

[0075] In one example, the VOC concentrator rotor 104 rotates at a rotational speed between 2 rotations per hour (rph) and 8 rph. The VOC concentrator rotor 104 can rotate at speeds other than these speeds without departing from the scope of this disclosure.

[0076] The desorption fan 116 is involved in the desorption process. In particular, the desorption fan 116 draws in the heated second portion 132 of the waste liquid 128 through the desorption zone 110 of the VOC concentrator rotor 104. Thus, the rotational speed of the desorption fan 116 affects the flow rate of the second portion 132 of the waste liquid 128 through the VOC concentrator rotor 104 at the desorption zone 110. Thus, the rotational speed of the desorption fan 116 can affect the overall efficiency of the VOC removal system 100, as will be described in more detail below. The desorption fan 116 includes a control circuit that enables the speed of the desorption fan 116 to be automatically adjusted by the control system 126, as will be described in more detail below.

[0077] The second portion 132 of the waste liquid 128 is transferred from the desorption fan 116 to the heat exchanger 118. The heat exchanger 118 heats the second portion 132 of the waste liquid 128.

[0078] The second portion 132 of the waste liquid 128 is transferred from the heat exchanger 118 to the burner 120. The burner 120 removes VOCs from the second portion 132 of the waste liquid 128. In particular, the burner 120 generates a high-temperature flame. The high-temperature flame causes oxidation of the VOCs. The oxidation of the VOCs causes the VOCs to be converted into H 2 O and CO 2Thus, at the burner 120, the hazardous VOCs are converted into harmless compounds. In this way, the burner 120 completes the removal of VOCs from the waste liquid 128. The VOCs are removed from the first portion 130 of the waste liquid 128 at the adsorption zone 106 of the VOC concentrator rotor 104. The VOCs are removed from the second portion 132 of the waste liquid 128 at the burner 120.

[0079] In one embodiment, the burner 120 operates at a temperature between 690 °C and 750 °C. The burner 120 can operate at other temperatures without departing from the scope of this disclosure. The temperature of the burner 120 can affect the overall VOC removal efficiency of the VOC removal system 100. Therefore, the burner 120 includes a control circuit that enables the temperature of the burner 120 to be controlled automatically by the control system 126.

[0080] The purified second portion 132 of the waste liquid 128 is transferred from the burner 120 to the heat exchanger 118. The heat exchanger 118 extracts heat from the purified second portion 132 to the waste liquid 128 and heats the second portion 132 of the waste liquid 128 from the desorption fan 116. This cools the purified second portion 132 of the waste liquid 128, but the purified second portion 132 of the waste liquid 128 remains hot after being transferred from the heat exchanger 118.

[0081] The purified second portion 132 of the waste liquid 128 is transferred from the heat exchanger 118 to the heat exchanger 112. The heat exchanger 112 extracts heat from the purified second portion 132 of the waste liquid 128 and heats the second portion 132 of the waste liquid 128 from the cooling zone 108. Therefore, the temperature imparted to the second portion 132 of the waste liquid 128 at the burner 120 is used to heat the second portion 132 of the waste liquid 128 that is about to enter the desorption zone 110. As previously described, the heat exchanger 112 can heat the VOC-containing second portion 132 to a temperature suitable for desorbing the VOCs onto the VOC concentrator rotor 104.

[0082] The purified second portion 132 of the waste liquid 128 is transferred from the heat exchanger 112 to the exhaust chimney 114. The exhaust chimney 114 transfers the purified second portion 132 of the waste liquid 128 to the atmosphere. In this way, the VOC removal system 100 removes VOCs from the waste liquid 128 of the semiconductor process 102.

[0083] The VOC removal system 100 includes a VOC sensor 134 and a VOC sensor 136. The VOC sensor 134 monitors the VOC concentration in the waste liquid 128 before any VOC is removed by the VOC removal system 100. The VOC sensor 136 monitors the VOC concentration in the waste liquid 128 after VOCs are removed from both the first portion 130 and the second portion 132 of the waste liquid 128. The VOC sensor 136 can be placed in the exhaust chimney 114, outside the exhaust chimney 114, or at another location that helps measure the VOC concentration in the purified waste liquid 128. The VOC sensors 134 and 136 provide VOC concentration signals or data to the control system 126. The control system 126 can calculate the total VOC removal efficiency based on the VOC concentration signals or data from the VOC sensors 134 and 136.

[0084] In one embodiment, the VOC sensors 134 and 136 are compound analyzers. A compound analyzer can detect the presence and concentration of selected compounds in a fluid. Thus, a compound analyzer can detect the presence and concentration of VOCs in the waste liquid 128. In one embodiment, there is only a single compound analyzer. The single compound analyzer can analyze fluid samples from the waste liquid 128 at both the inlet and the outlet of the VOC removal system 100.

[0085] The temperature sensors 122 and 124 cooperate to measure the temperature of the VOC concentrator rotor 104. The temperature of the VOC concentrator rotor 104 can affect the total VOC removal efficiency of the VOC removal system 100. Each of the temperature sensors 122 generates a sensor signal indicating the temperature of the VOC concentrator rotor 104. The temperature signals from the temperature sensors 122 and 124 can be used to determine the temperature of the VOC concentrator rotor 104.

[0086] In one embodiment, the temperature sensor 122 is placed on the input side of the adsorption zone 106. The temperature sensor 124 is placed on the output side of the adsorption zone 106. There can be a temperature gradient across the VOC concentrator rotor 104. Thus, the temperature signals from the temperature sensors 122 and 124 can be used to determine the temperature gradient or the average temperature of the VOC concentrator rotor 104. The VOC removal system 100 can include additional temperature sensors that are located at various positions to sense the temperature of the VOC concentrator rotor 104. All the temperature signals can be provided to the control system 126. In one embodiment, the temperature sensor can be positioned to measure the temperature of the second portion 132 of the waste liquid 128 between the heat exchanger 112 and the desorption zone 110.

[0087] In some cases, components of the VOC removal system 100 may be located in positions that are subject to high temperature fluctuations, air pressure fluctuations, or humidity fluctuations. For example, in some cases, the VOC concentrator rotor 104 may be located on the roof of a semiconductor manufacturing facility. In such cases, the time of day, the current weather, and other factors may affect the temperature of the VOC concentrator rotor 104, which in turn may affect the overall VOC removal efficiency of the VOC removal system 100.

[0088] The control system 126 receives data related to various parameters of the VOC removal system 100. The control system 126 receives temperature signals from the temperature sensors 122 and 124. The control system 126 receives VOC concentration signals from the VOC sensors 134 and 136. The control system 126 receives data indicating the rotational speed of the VOC concentrator rotor 104, the rotational speed of the desorption fan 116, the temperature of the burner 120, and the heat transfer parameters of the heat exchangers 112 and 118. Thus, the control system 126 is aware of the various current operating parameters of the VOC removal system 100.

[0089] In one embodiment, the control system 126 includes an analysis model trained by machine learning. The analysis model trained by machine learning is trained by a machine learning process to predict the VOC removal efficiency in a future time period based on the current operating parameters of the VOC removal system 100, and to generate recommended operating parameters that can be implemented to improve the VOC removal efficiency.

[0090] Figure 2 is a block diagram of the Figure 1 control system 126 according to one embodiment. According to one embodiment, the control system 126 is configured to control the operation of the VOC removal system 100. The control system 126 uses machine learning to adjust the parameters of the VOC removal system 100. The control system 126 may adjust the parameters of the VOC removal system 100 to maintain a high VOC removal efficiency.

[0091] In one embodiment, the control system 126 includes an analysis model 140 and a training module 141. The training module 141 trains the analysis model 140 by a machine learning process. The machine learning process trains the analysis model 140 to predict the future VOC removal efficiency and to select parameters for the VOC removal process that will result in a high VOC removal efficiency. Although the illustrated training module 141 is separate from the analysis model 140, in practice, the training module 141 may be part of the analysis model 140.

[0092] The control system 126 includes or stores training set data 142. The training set data 142 includes historical VOC removal efficiency data 144 and historical VOC removal condition data 146. The historical VOC removal efficiency data 144 includes VOC removal efficiency data for the VOC removal process. The historical VOC removal condition data 146 includes data related to process conditions or parameters during the VOC removal process associated with the historical VOC removal efficiency data 144. As will be elaborated in more detail below, the training module 141 uses the historical VOC removal efficiency data 144 and the historical VOC removal condition data 146 to train the analysis model 140 in a machine learning process.

[0093] In one embodiment, the historical VOC removal efficiency data 144 includes data indicating the efficiency of the VOC removal process. For example, during the operation of a semiconductor manufacturing facility, thousands or millions of semiconductor wafers may be processed over the course of several months or years. A corresponding large number of VOC-generating semiconductor processes are performed when processing the wafers. The VOC removal process is performed to remove VOCs from the waste liquid generated by the semiconductor process. The historical VOC removal efficiency data 144 includes the VOC removal efficiency of these VOC removal processes or the time periods during which the VOC removal is performed for a selected period.

[0094] In one embodiment, the historical VOC removal condition data 146 includes various process conditions or parameters during the VOC removal process associated with the historical VOC removal efficiency data 144. Thus, for each VOC removal efficiency value in the historical VOC removal efficiency data 144, the historical VOC removal condition data 146 may include the process conditions or parameters that existed during the time period associated with the VOC removal efficiency value. The historical VOC removal condition data 146 may include the rotational speed of the VOC concentrator rotor 104, the temperature of the VOC concentrator rotor 104, the rotational speed of the desorption fan 116, the temperature of the burner 120, the temperature of the VOC concentrator rotor 104 in the adsorption zone 106, the temperature of the VOC concentrator rotor 104 in the desorption zone 110, the inlet VOC concentration of the waste liquid 128, the outlet VOC concentration of the waste liquid 128, the quantity of the waste liquid 128, or other parameters of the VOC removal process.

[0095] In some embodiments, the training set data 142 links historical VOC removal efficiency data 144 with historical VOC removal condition data 146. In other words, each VOC removal efficiency value in the historical VOC removal efficiency data 144 is linked to the process condition data associated with the VOC removal process corresponding to the VOC removal efficiency value. In this way, the historical VOC removal efficiency values are labels for the machine learning process. As will be elaborated in more detail below, the labeled training set data 142 can be used in the machine learning process to train the analysis model 140 to predict future VOC removal efficiency and generate recommended VOC removal parameters to improve future VOC removal efficiency.

[0096] In one embodiment, the analysis model 140 includes a neural network. The training of the analysis model 140 will be described in connection with the neural network. However, other types of analysis models or algorithms can be used without departing from the scope of this disclosure. The training module 141 uses the training set data 142 to train the neural network in a machine learning process. During the training process, the neural network receives the historical VOC removal condition data 146 from the training set data 142 as input. During the training process, the neural network outputs predicted VOC removal efficiency data. The predicted VOC removal efficiency data predicts the VOC removal efficiency that will be caused by the historical VOC removal condition data 146. The training process trains the neural network to generate the predicted VOC removal efficiency data. The training process also trains the neural network to generate recommended VOC removal parameters to improve the VOC removal efficiency.

[0097] In one embodiment, the neural network includes multiple neural layers. The various neural layers include neurons that define one or more internal functions. The internal functions are based on the weighted values associated with the neurons of each neural layer of the neural network. During training, for each historical VOC removal condition data 146, the control system 126 compares the predicted VOC removal efficiency data with the actual historical VOC removal efficiency data 144 associated with those process conditions. The control system 126 generates an error function that indicates how closely the predicted VOC removal efficiency data matches the historical VOC removal efficiency data 144. The control system 126 then adjusts the internal functions of the neural network. Since the neural network generates the predicted VOC removal efficiency data based on the internal functions, adjusting the internal functions will result in different predicted VOC removal efficiency data for the same historical VOC removal condition data 146. Adjusting the internal functions can result in predicted VOC removal efficiency data that produces a larger error function (matches the historical VOC removal efficiency data 144 worse) or a smaller error function (matches the historical VOC removal efficiency data 144 better).

[0098] After adjusting the internal function of the neural network, the historical VOC removal condition data 146 is passed to the neural network again, and the analysis model 140 generates the predicted VOC removal efficiency data again. The training module 141 compares the predicted VOC removal efficiency data with the historical VOC removal efficiency data 144 again. The training module 141 adjusts the internal function of the neural network again. This process is repeated in many iterations of monitoring the error function and adjusting the internal function of the neural network until the set of internal functions that results in the predicted VOC removal efficiency data matching the historical VOC removal efficiency data 144 is found across the entire training set.

[0099] At the start of the training process, the predicted VOC removal efficiency data is likely to not match the historical VOC removal efficiency data 144 very closely. However, as the training process progresses through many iterations of adjusting the internal function of the neural network, the trend of the error function will become smaller and smaller until the set of internal functions that results in the predicted VOC removal efficiency data matching the historical VOC removal efficiency data 144 is found. The identification of the set of internal functions that results in the predicted VOC removal efficiency data matching the historical VOC removal efficiency data 144 corresponds to the completion of the training process.

[0100] In one embodiment, the analysis model 140 includes two neural networks coupled together in an encoder-decoder configuration. The encoder neural network is trained using the above training process to generate the predicted VOC removal efficiency. The decoder network is trained to receive the predicted VOC removal efficiency and reproduce the historical VOC removal condition data 146 that led to the predicted VOC removal efficiency.

[0101] The training of the decoder neural network is similar to the training of the encoder neural network. The decoder neural network includes multiple neural layers as described above in connection with the encoder neural network. The decoder neural network receives the VOC removal efficiency value as input and generates the historical VOC removal condition as output. The training process uses the historical VOC removal condition data 146 as the label. For each VOC removal efficiency value, the decoder neural network generates the predicted VOC removal condition. The predicted VOC removal condition is compared with the historical VOC removal condition data 146 and an error function is generated. The internal function of the decoder neural network is adjusted in multiple iterations until the decoder neural network can generate the predicted VOC removal condition data that matches the historical VOC removal condition data 146 within the error tolerance.

[0102] In one embodiment, after the trained analysis model 140, the analysis model 140 can be used to generate recommended process conditions that will result in improved VOC removal efficiency. For example, the current VOC removal process conditions or parameters are provided to the encoder neural network of the analysis model 140. The encoder neural network generates a predicted future VOC removal efficiency based on the current VOC removal process conditions or parameters. If the predicted future VOC removal efficiency is lower than a selected threshold value, a higher VOC removal efficiency value can be provided to the decoder neural network. The decoder neural network then generates a recommended set of VOC process removal parameters that will result in a higher VOC removal efficiency. The control system 126 can then adjust the operation of the various components of the VOC removal system 100 to implement the recommended VOC process removal parameters.

[0103] In one embodiment, the control system 126 includes processing resources 148, memory resources 150, and communication resources 152. The processing resources 148 can include one or more controllers or processors. The processing resources 148 are configured to execute software instructions, process data, make VOC parameter control decisions, perform signal processing, read data from memory, write data to memory, and perform other processing operations. The processing resources 148 can include physical processing resources 148 located at the site or facility of the VOC removal system 100. The processing resources can include virtual processing resources 148 remote from the on-site VOC removal system 100 or the facility where the VOC removal system 100 is located. The processing resources 148 can include cloud-based processing resources that include processors and servers accessed via one or more cloud computing platforms.

[0104] In one embodiment, the memory resources 150 can include one or more computer-readable memories. The memory resources 150 are configured to store software instructions associated with the functions of the control system and its components, including but not limited to the analysis model 140. The memory resources 150 can store data associated with the functions of the control system 126 and its components. The data can include training set data 142, current process condition data, and any other data associated with the operation of the control system 126 and its components. The memory resources 150 can include physical memory resources located at the site or facility of the VOC removal system 100. The memory resources can include virtual memory resources remote from the site or facility of the VOC removal system 100. The memory resources 150 can include cloud-based memory resources accessed via one or more cloud computing platforms.

[0105] In one embodiment, the communication resources may include resources that enable the control system 126 to communicate with components associated with the VOC removal system 100. For example, the communication resources 152 may include wired and wireless communication resources that enable the control system 126 to receive sensor data associated with the VOC removal system 100 and control the devices of the VOC removal system 100. The communication resources 152 may enable the control system 126 to control the operation of the burner 120, control the rotational speed of the desorption fan 116, control the rotational speed of the VOC concentrator rotor 104, or control other aspects of the components of the VOC removal system 100. The communication resources 152 may enable the control system 126 to communicate with a remote system. The communication resources 152 may include or facilitate communication via one or more networks, such as a wired network, a wireless network, the Internet, or an internal network. The communication resources 152 may enable the components of the control system 126 to communicate with each other.

[0106] In one embodiment, the analysis model 140 is implemented via the processing resources 148, the memory resources 150, and the communication resources 152. The control system 126 may be a distributed control system, with its components and resources located remotely from each other and from the VOC removal system 100.

[0107] In one embodiment, the control system 126 includes a data integration server 153. The data integration server 153 collects data to be provided to the analysis model 140. For example, the data integration server 153 may collect and store training set data 142. The data integration server 153 may receive data from one or more controllers implemented according to the processing resources 148 and the memory resources 150. The one or more controllers may provide data related to the current operating parameters of the VOC removal system 100 to the data integration server. The one or more controllers may also provide data related to the inlet VOC concentration, the combined VOC concentration, the current VOC removal efficiency, and the waste liquid flow rate to the data integration server. The data integration server 153 may format the data such that it can be processed by the analysis model 140.

[0108] The data integration server 153 may also receive data from the analysis model 140. For example, the data integration server 153 may receive the predicted removal efficiency and the recommended parameter adjustments from the analysis model 140. The data integration server 153 may provide this data to one or more controllers configured to control the parameters of the VOC removal system 100. The data integration server 153 is implemented according to the processing resources 148, the memory resources 150, and the communication resources 152.

[0109] Figure 3 is according to one embodiment Figure 2Block diagram of the analysis model 140, which illustrates the operating and training aspects of the analysis model 140. The analysis model 140 includes an encoder neural network 160 and a decoder neural network 162, as described in connection with Figure 2 as described.

[0110] As previously described, the training set data 142 includes data related to a plurality of previously performed VOC removal processes. Each previously performed VOC removal process occurs under specific process conditions and results in a specific VOC removal efficiency. The process conditions for each VOC removal efficiency value are formatted into individual process condition vectors 164. The process condition vector 164 includes a plurality of data fields 166. Each data field 166 corresponds to a specific process condition.

[0111] Figure 3 An example of illustrates a single process condition vector 164 that will be passed to the encoder 160 of the analysis model 140 during the training process. In Figure 3 the example, the process condition vector 164 includes nine data fields 166. The first data field 166 corresponds to the rotational speed of the VOC concentrator rotor 104. The second data field 166 corresponds to the rotational speed of the desorption fan 116. The third data field 166 corresponds to the temperature of the burner 120. The fourth data field 166 corresponds to the temperature of the VOC concentrator rotor 104 at the desorption zone 110. The fifth data field 166 corresponds to the temperature of the VOC concentrator rotor 104 at the adsorption zone 106. The sixth data field 166 corresponds to the inlet VOC concentration in the waste liquid 128. The seventh data field 166 corresponds to the outlet VOC concentration in the waste liquid 128. The eighth data field 166 corresponds to the opening degree of the air flow valve of the burner 120. The ninth data field corresponds to the total flow rate of the waste liquid 128. In practice, each process condition vector 164 may include more or fewer data fields than those shown in Figure 3 without departing from the scope of the present disclosure. Each process condition vector 164 may include different types of process conditions without departing from the scope of the present disclosure. Figure 3 The specific process conditions illustrated in are given only as examples. Each process condition is represented by a numerical value in the corresponding data field 166.

[0112] The encoder 160 includes a plurality of neural layers 168a - 168c. Each neural layer includes a plurality of nodes 170. Each node 170 may also be referred to as a neuron. Each node 170 from the first neural layer 168a receives data values from each data field of the process condition vector 164. Thus, in Figure 3 the example, each node 170 from the first neural layer 168a receives nine data values because the process condition vector 164 has nine data fields. Each neuron 170 includes in Figure 3Individually labeled internal mathematical function F(x). Each node 170 of the first neural layer 168a generates a scalar value by applying the internal mathematical function F(x) to the data value of the data field 166 from the process condition vector 164. Additional details regarding the internal mathematical function F(x) are provided below.

[0113] In Figure 3 the example, each neural layer 168a - 168e in both the encoder 160 and the decoder 162 is a fully connected layer. This means that each neural layer has the same number of nodes as the next neural layer. In Figure 3 the example, each neural layer 168a - 168e contains five nodes. However, the neural layers of the encoder 160 and the decoder 162 can contain a different number of layers than Figure 3 shown without departing from the scope of this disclosure.

[0114] Each node 170 of the second neural layer 168b receives the scalar value generated by each node 170 of the first neural layer 168a. Thus, in Figure 3 the example, each node 170 of the second neural layer 168b receives five scalar values because there are five nodes 170 in the first neural layer 168a. Each node 170 of the second neural layer 168b generates a scalar value by applying the individual internal mathematical function F(x) to the scalar values from the first neural layer 168a.

[0115] There may be one or more additional neural layers between the neural layer 168b and the neural layer 168c. The last neural layer 168c of the encoder 160 receives five scalar values from five nodes of the previous neural layer (not shown). The output of the last neural layer is the predicted VOC removal efficiency 172.

[0116] During the machine learning process, the analysis module 140 compares the predicted VOC removal efficiency data value 172 with the actual removal efficiency value. The analysis model 140 generates an error value that indicates the error or difference between the predicted VOC removal efficiency from the data value 172 and the actual removal efficiency. The error value is used to train the encoder 160.

[0117] By discussing the internal mathematical function F(x), the training of the encoder 160 can be more comprehensively understood. Although all nodes 170 are labeled with the internal mathematical function F(x), the mathematical function F(x) for each node is unique. In one example, each internal mathematical function has the following form:

[0118] F(x) = x 1 *w 1 +x 2 *w 2 +…x n *w1 +b

[0119] In the above equation, each value x 1 -x n corresponds to the data value received from node 170 in the previous neural layer, or in the case of the first neural layer 168a, each value x 1 -x n corresponds to the individual data values of data fields 166 from the process condition vector 164. Thus, n for a certain node is equal to the number of nodes in the previous neural layer. The value w 1 -w n is a scalar weighting value associated with the corresponding node from the previous layer. The analysis model 140 selects the weighting value w 1 -w n . The constant b is a scalar bias value and can also be multiplied by the weighting value. The value generated by node 170 is based on the weighting value w 1 -w n . Thus, each node 170 has n weighting values w 1 -w n . Although not shown above, each function F(x) can also include an activation function. Multiply the sum set forth in the above equation by the activation function. Examples of activation functions can include the rectified linear unit (ReLU) function, sigmoid function, hyperbolic tangent function, or other types of activation functions. Each function F(x) can also include a transfer function.

[0120] After the error value has been calculated, the analysis model 140 adjusts the weighting values w for the various nodes 170 of the various neural layers 168a - 168c 1 -w n . After the analysis model 140 adjusts the weighting value w 1 -w n , the analysis model 140 provides the process condition vector 164 to the input neural layer 168a again. Since the weighting values are different for the various nodes 170 of the analysis model 140, the predicted VOC removal efficiency 172 will be different from that in the previous iteration. The analysis model 140 generates an error value again by comparing the actual removal efficiency with the predicted VOC removal efficiency 172.

[0121] The analysis model 140 adjusts the weighting values w associated with the various nodes 170 again 1 -w n . The analysis model 140 processes the process condition vector 164 again and generates the predicted VOC removal efficiency 172 and the associated error value. The training process includes adjusting the weighting value w in multiple iterations 1 -w nuntil the error value is minimized.

[0122] Figure 3 Illustratively, a single process condition vector 164 is passed to the encoder 160. In practice, the training process involves passing a large number of process condition vectors 164 through the analysis model 140, generating a predicted VOC removal efficiency 172 for each process condition vector 164, and generating an associated error value for each predicted VOC removal efficiency. The training process may also include generating a summary error value that indicates the average error of all predicted VOC removal efficiencies for a batch of process condition vectors 164. After processing each batch of process condition vectors 164, the analysis model 140 adjusts the weighting value w 1 -w n . The training process continues until the average error across all process condition vectors 164 is less than a selected threshold tolerance. When the average error is less than the selected threshold tolerance, the training of the encoder 160 is complete, and the analysis model 140 is trained to accurately predict the VOC removal efficiency based on the process conditions.

[0123] The decoder 162 operates and is trained in a manner similar to the encoder 160 described above. During the training process of the decoder 162, the decoder 162 receives the VOC removal efficiency values associated with the process condition vectors 164. The VOC removal efficiency values are received by each node 170 of the first neural layer 168d of the decoder 162. The nodes 170 of the first neural layer 168d apply their individual functions F(x) to the VOC removal efficiency values and pass the resulting scalar values to the nodes 170 of the next neural layer 168e. After the last neural layer 168f processes the scalar values received from the previous neural layer (not shown), the last neural layer 168f outputs a predicted process condition vector 174. The predicted process condition vector 174 has the same form as the process condition vector 164. The data fields 175 of the predicted process condition vector 174 represent the same parameters or conditions as the data fields 166 of the process condition vector 164.

[0124] The training process compares the predicted process condition vector 174 with the process condition vector 164 and determines an error value. The waiting parameters of the function F(x) of the nodes 170 of the decoder 162 are adjusted and the removal efficiency values are provided to the decoder 162 again. The decoder 162 generates a predicted process condition vector 174 again and determines an error value. This process is repeated for all process condition vectors 164 in the historical VOC removal condition data 146 and for all historical VOC removal efficiency values from the historical VOC removal efficiency data 144 until the decoder 162 can generate a predicted process condition vector 174 that matches the corresponding process condition vector 164 for each historical VOC removal efficiency value. When the predicted cumulative error value is below the threshold error value, the training process is complete.

[0125] After the encoder 160 and the decoder 162 have been trained as described above, the analysis model 140 is ready to generate recommended VOC removal parameters to improve the VOC removal efficiency of the VOC removal system 100. During operation, the analysis model 140 receives a current process condition vector representing the current conditions or parameters of the VOC removal system 100. The encoder 160 processes the current process condition vector and generates a predicted future VOC removal efficiency based on the current process condition vector. If the predicted future VOC removal efficiency is less than a threshold removal efficiency, the decoder 162 is used to generate a set of recommended process conditions that will result in a higher VOC removal efficiency. In practice, the decoder 162 receives an increased removal efficiency data value that is higher than the predicted future removal efficiency value generated by the encoder 160. The decoder 162 then generates a predicted process condition vector 174 based on the higher removal efficiency data value.

[0126] The predicted process condition vector 174 contains recommended process condition values for certain process condition types. For example, the predicted process condition vector 174 may contain a recommended VOC concentrator rotor speed, a recommended desorption fan speed, and a recommended burner temperature. The control system 126 can then control the corresponding components of the VOC removal system 100 to implement the recommended process conditions.

[0127] In one embodiment, the control system 126 applies constraint conditions to the recommended process condition values. For example, the control system 126 can enforce the following constraint conditions: any recommended adjustment must fall within a predetermined difference value of the current value of the parameter. For example, the control system 126 can enforce the following constraint conditions: the difference between the recommended burner temperature and the current burner temperature does not exceed 2 °C, the difference between the recommended VOC concentrator rotor speed and the current VOC concentrator rotor speed does not exceed 1 rph, the difference between the recommended desorption fan speed and the current desorption fan speed does not exceed 1 Hz, and the difference between the recommended desorption gas temperature and the current desorption gas temperature does not exceed 2 °C. Maximum deviation values other than those described above can be utilized without departing from the scope of this disclosure.

[0128] In addition, the control system 126 can enforce constraint conditions on the magnitude of the values of the recommended process parameters. For example, the control system 126 can apply the following constraint conditions: the burner temperature should be maintained between 710 °C and 730 °C, the desorption gas temperature should be between 190 °C and 210 °C, the desorption fan speed should be between 40 Hz and 50 Hz, and the concentrator rotor speed should be between 2 rph and 10 rph. Values other than those described above can be utilized without departing from the scope of this disclosure.

[0129] A particular instance of the neural network-based analysis model 140 has been described in connection with Figure 3However, other types of analytical models based on neural networks, or types of analytical models other than neural networks, may be utilized without departing from the scope of the present disclosure. Additionally, a neural network may have a different number of neural layers, which have a different number of nodes, without departing from the scope of the present disclosure.

[0130] Figure 4 is a flowchart of process 400 for accurately predicting future VOC removal efficiency according to an embodiment using an encoder of a training analytical model (such as Figure 2 and Figure 3 analytical model 140). The various steps of process 400 may utilize the components, processes, and techniques described in connection with Figures 1 to 3 . Accordingly, Figure 4 is described with reference to Figures 1 to 3 .

[0131] At step 402, process 400 collects training set data 142 that includes historical VOC removal efficiency data 144 and historical VOC removal condition data 146. This may be achieved by using a data mining system or process. The data mining system or process may collect the training set data 142 by accessing one or more databases associated with the VOC removal system 100 and collecting and organizing the various types of data contained in the one or more databases. The data mining system or process, or another system or process, may process and format the collected data to produce the training set data 142. The training set data 142 may include historical VOC removal efficiency data 144 and historical VOC removal condition data 146, as described in connection with Figure 2 .

[0132] At step 404, process 400 inputs the historical VOC removal condition data 146 into the encoder 160. In one example, this may include inputting the historical VOC removal condition data 146 into the analytical model 140 having a training module 141, as described in connection with Figure 2 and Figure 3 . The historical VOC removal condition data 146 may be provided to the encoder 160 in a sequential discrete set. The historical VOC removal condition data 146 may be provided to the encoder 160 as a vector. Each set may include one or more vectors that are formatted for receipt and processing by the encoder 160. The historical VOC removal condition data 146 may be provided to the encoder 160 in other formats without departing from the scope of the present disclosure.

[0133] At step 406, process 400 generates predicted VOC removal efficiency data based on the historical VOC removal condition data 146. In particular, the analytical model 140 generates predicted VOC removal efficiency data for each historical VOC removal condition data 146.

[0134] At step 408, the predicted VOC removal efficiency data is compared with the historical VOC removal efficiency data 144. In particular, the predicted VOC removal efficiency data for each historical VOC removal condition data 146 is compared with the historical VOC removal efficiency data 144 associated with this historical VOC removal condition data 146. The comparison may result in an error function that indicates how closely the predicted VOC removal efficiency data matches the historical VOC removal efficiency data 144. This comparison is performed for each set of predicted VOC removal efficiency data. In one embodiment, this process may include generating a summary error function or indication that indicates how all of the predicted VOC removal efficiency data compares to the historical VOC removal efficiency data 144. These comparisons may be performed by the training module 141 or by the analysis model 140. The comparison may include types of functions or data other than those described above without departing from the scope of this disclosure.

[0135] At step 410, process 400 determines whether the predicted VOC removal efficiency data matches the historical VOC removal efficiency data 144 based on the comparison generated at step 408. In one instance, if the summary error function is greater than the error tolerance, process 400 determines that the predicted VOC removal efficiency data does not match the historical VOC removal efficiency data 144. In one instance, if the summary error function is less than the error tolerance, process 400 determines that the predicted VOC removal efficiency data matches the historical VOC removal efficiency data 144.

[0136] In one embodiment, if the predicted VOC removal efficiency data does not match the historical VOC removal efficiency data 144 at step 410, process 400 proceeds to step 412. At step 412, process 400 adjusts the internal functions associated with the analysis model 140. In one instance, the training module 141 adjusts the internal functions associated with the encoder 160. Process 400 returns from step 412 to step 404. At step 404, the historical VOC removal condition data 146 is provided to the analysis model 140 again. Because the internal functions of the encoder 160 have been adjusted, the analysis model 140 will generate predicted VOC removal efficiency data different from that in the previous cycle. Process 400 proceeds to steps 406, 408, and 410 and calculates the summary error. If the predicted VOC removal efficiency data does not match the historical VOC removal efficiency data 144, process 400 returns to step 412 and adjusts the internal functions of the encoder 160 again. This process is performed in multiple iterations until the encoder 160 generates predicted VOC removal efficiency data that matches the historical VOC removal efficiency data 144.

[0137] In one embodiment, if the predicted VOC removal efficiency data at step 410 matches the historical VOC removal efficiency data 144, then process step 410 in process 400 proceeds to step 414. At step 414, training is complete. The encoder 160 of the analysis model 140 is now ready for predicting VOC removal efficiency. Process 400 may include other steps or configurations in addition to those shown and described herein without departing from the scope of the disclosure.

[0138] Figure 5 FIG. 500 is a flow chart of a process 500 for training a decoder of an analysis model to identify process conditions that will result in effective VOC removal from the waste liquid 128 according to one embodiment. The various steps of process 500 may utilize the components, processes, and techniques described in conjunction with Figures 1 to 4 Thus, Figure 5 is described with reference to Figures 1 to 4 to describe.

[0139] At step 502, process 500 collects training set data 142 that includes historical VOC removal efficiency data 144 and historical VOC removal condition data 146. This may be accomplished by using a data mining system or process. The data mining system or process may collect the training set data 142 by accessing one or more databases associated with the VOC removal system 100 and collecting and organizing the various types of data contained in the one or more databases. The data mining system or process or another system or process may process and format the collected data to produce the training set data 142. The training set data 142 may include historical VOC removal efficiency data 144 and historical VOC removal condition data 146, as described in conjunction with Figure 2 described.

[0140] At step 504, process 500 inputs the historical VOC removal efficiency data 144 into the decoder 162. The historical VOC removal efficiency data 144 may be provided to the decoder 162 in consecutive discrete values.

[0141] At step 506, process 500 generates predicted VOC removal condition data based on the historical VOC removal condition data 146. In particular, the decoder 162 generates predicted VOC removal condition data for each historical VOC removal efficiency value.

[0142] At step 508, the predicted VOC removal condition data is compared with the historical VOC removal condition data 146. In particular, the predicted VOC removal condition data for each historical VOC removal efficiency value is compared with the historical VOC removal condition data 146 associated with that historical VOC removal efficiency value. The comparison can result in an error function that indicates how closely the predicted VOC removal condition data matches the historical VOC removal condition data 146. This comparison is performed for each set of predicted VOC removal condition data. In one embodiment, this process can include generating a summary error function or indication that indicates how all of the predicted VOC removal condition data compares to the historical VOC removal condition data 146. These comparisons can be performed by the training module 141 or by the analysis model 140. The comparison can include types of functions or data other than those described above without departing from the scope of this disclosure.

[0143] At step 510, process 500 determines whether the predicted VOC removal condition data matches the historical VOC removal condition data 146 based on the comparison generated at step 508. In one example, if the summary error function is greater than the error tolerance, process 500 determines that the predicted VOC removal condition data does not match the historical VOC removal condition data 146. In one example, if the summary error function is less than the error tolerance, process 500 determines that the predicted VOC removal data matches the historical VOC removal condition data 146.

[0144] In one embodiment, if the predicted VOC removal efficiency data does not match the historical VOC removal efficiency data 144 at step 510, process 500 proceeds to step 512. At step 512, process 500 adjusts the internal functions associated with the analysis model 140. In one example, the training module 141 adjusts the internal functions associated with the decoder 162. Process 500 returns from step 512 to step 504, and proceeds to steps 506, 508, and 510 and calculates the summary error. If the predicted VOC removal condition data does not match the historical VOC removal condition data 146, process 500 returns to step 512 and adjusts the internal functions of the decoder 162 again. This process is performed in multiple iterations until the decoder 162 generates predicted VOC removal condition data that matches the historical VOC removal condition data 146.

[0145] In one embodiment, if the predicted VOC removal condition data matches the historical VOC removal condition data 146, process 500 proceeds to step 514. At step 514, the training of decoder 162 is completed. Analysis model 140 is now ready to be used to identify process conditions for VOC removal system 100. Process 500 may include other steps or configurations in addition to those shown and described herein without departing from the scope of this disclosure.

[0146] Figure 6 is a flowchart of method 600 for operating a VOC removal system according to one embodiment. The various steps of method 600 may utilize the components, processes, and techniques described in connection with Figures 1 to 5 In step 602, method 600 includes removing VOCs from the waste liquid of a semiconductor process using a VOC removal system. An example of a VOC removal system is Figure 1 the VOC removal system of. In step 604, method 600 includes providing the parameters of the VOC removal system to an analysis model trained by a machine learning process. An example of an analysis model is Figure 2 analysis model 140 of. In step 606, method 600 includes generating a predicted future VOC removal efficiency by the analysis model based on the parameters. In step 608, method 600 includes generating adjusted parameters by the analysis model based on the predicted future VOC removal efficiency and the parameters. In step 610, method 600 includes adjusting the VOC removal system based on the adjusted parameters.

[0147] Figure 7 is a flowchart of method 700 for operating a VOC removal system according to one embodiment. The various steps of method 700 may utilize the components, processes, and techniques described in connection with Figures 1 to 6 In step 702, method 700 includes transferring the waste liquid from a semiconductor process to a VOC concentrator rotor. An example of a semiconductor process is Figure 1 the semiconductor process 102 of. An example of a VOC concentrator rotor is Figure 1 the VOC concentrator rotor 104 of. In step 704, method 700 includes adsorbing VOCs from a first portion of the waste liquid using the VOC concentrator rotor. In step 706, method 700 includes transferring a second portion of the waste liquid from the VOC concentrator rotor to a heater. In step 708, method 700 includes transferring the second portion of the waste liquid from the heater to the VOC concentrator rotor. An example of a heater is Figure 1 the heat exchanger 112 of. In step 710, method 700 includes desorbing the VOCs from the VOC concentrator rotor into the second portion of the waste liquid by operating a desorption fan adjacent to the VOC concentrator rotor. An example of a desorption fan is Figure 1The desorption fan 116. At step 712, method 700 includes burning VOCs from the second portion of the waste liquid using a burner. An example of the burner is Figure 1 The burner 120. At step 714, method 700 includes measuring the temperature of the VOC concentrator rotor. At step 716, method 700 includes generating a predicted future VOC removal efficiency at least partially based on the temperature of the VOC concentrator rotor using an analysis model trained through a machine learning process. An example of the analysis model is Figure 2 The analysis model 140. At step 718, method 700 includes generating adjustment parameters using the analysis model based on the predicted future VOC removal efficiency and the temperature of the VOC concentrator rotor. At step 720, method 700 includes adjusting the rotational speed of the VOC concentrator rotor and the temperature of the burner based on the adjustment parameters.

[0148] In one embodiment, the method includes removing VOCs from the waste liquid of a semiconductor process using a VOC removal system and providing the parameters of the VOC removal system to an analysis model trained through a machine learning process. The method includes generating a predicted future VOC removal efficiency using the analysis model based on the parameters, generating adjustment parameters using the analysis model based on the predicted future VOC removal efficiency, and adjusting the VOC removal system based on the adjustment parameters.

[0149] In one embodiment, the step of generating the adjustment parameters includes: using the analysis model to identify changes to the VOC removal system that are predicted to result in improved VOC removal efficiency.

[0150] In one embodiment, the above parameters include a current VOC removal efficiency of the VOC removal system.

[0151] In one embodiment, the above parameters include the temperature and rotational speed of the VOC concentrator rotor.

[0152] In one embodiment, the above parameters include the temperature of the VOC burner.

[0153] In one embodiment, the above parameters include the temperature and speed of the desorption fan.

[0154] In one embodiment, the above adjustment parameters include a recommended rotational speed of the VOC concentrator rotor.

[0155] In one embodiment, the above adjustment parameters include a recommended rotational speed of the desorption fan.

[0156] In one embodiment, the above adjustment parameters include a recommended temperature of the VOC burner.

[0157] In one embodiment, the above adjustment parameters include a recommended desorption temperature.

[0158] In one embodiment, the VOC removal system includes a VOC concentrator rotor configured to adsorb VOCs from a first portion of the waste liquid from a semiconductor process and desorb the VOCs to a second portion of the waste liquid. The system includes: a desorption fan configured to drive the second portion of the waste liquid through a desorption zone of the VOC concentrator rotor; and a burner configured to burn the VOCs from the second portion of the waste liquid. The system includes a control system that includes an analysis model trained through a machine learning process and configured to analyze parameters of the VOC concentrator rotor, the desorption fan, and the burner, and configured to generate a predicted future VOC removal efficiency based on the parameters and generate adjustment parameters based on the parameters and the predicted future VOC removal efficiency. The control system is configured to adjust the operations of the VOC concentrator rotor, the desorption fan, and the burner based on the adjustment parameters.

[0159] In one embodiment, the above analysis model includes an encoder and a decoder. The encoder is configured to generate a predicted future VOC removal efficiency. The decoder is configured to generate adjustment parameters.

[0160] In one embodiment, the above VOC removal system further includes a first temperature sensor positioned adjacent to the VOC concentrator rotor and configured to sense the temperature of the VOC concentrator rotor, wherein the parameters include the temperature of the VOC concentrator rotor.

[0161] In one embodiment, the above VOC removal system further includes a second temperature sensor positioned adjacent to the upstream side of the VOC concentrator rotor and the first temperature sensor positioned adjacent to the downstream side of the VOC concentrator rotor, wherein the parameters include the temperature sensed by the first temperature sensor and the temperature sensed by the second temperature sensor.

[0162] In one embodiment, the above VOC removal system further includes a heat exchanger configured to receive heat from the second portion of the waste liquid downstream of the burner and supply the heat to the second portion of the waste liquid upstream of the desorption zone of the VOC concentrator rotor.

[0163] In one embodiment, the above VOC removal system further includes a VOC sensor configured to sense the inlet VOCs in the waste liquid.

[0164] In one embodiment, the above VOC removal system further includes a VOC sensor configured to sense the outlet VOCs in the waste liquid.

[0165] In one embodiment, the method includes transferring waste liquid from a semiconductor process to a VOC concentrator rotor, adsorbing VOCs from a first portion of the waste liquid using the VOC concentrator rotor, and transferring a second portion of the waste liquid from the VOC concentrator rotor to a heater. The method includes transferring the second portion of the waste liquid from the heater to the VOC concentrator rotor, desorbing VOCs from the VOC concentrator rotor to the second portion of the waste liquid by operating a desorption fan adjacent to the VOC concentrator rotor, and burning the VOCs from the second portion of the waste liquid using a burner. The method includes measuring the temperature of the VOC concentrator rotor and generating a predicted future VOC removal efficiency at least in part based on the temperature of the VOC concentrator rotor using an analysis model trained by a machine learning process. The method includes generating an adjustment parameter based on the predicted future VOC removal efficiency and the temperature of the VOC concentrator rotor using the analysis model and adjusting the speed of the VOC concentrator rotor and the temperature of the burner based on the adjustment parameter.

[0166] In one embodiment, the above method further includes: adjusting the rotational speed of the desorption fan based on the adjustment parameter.

[0167] In one embodiment, the above method further includes: measuring the current VOC removal efficiency and generating an adjustment parameter based on the current VOC removal efficiency.

[0168] In some embodiments according to the present disclosure, a method for removing volatile organic compounds (VOCs) includes: using a VOC concentration sensor to measure the VOC concentration of the exhaust gas of a semiconductor process; after measuring the VOC concentration of the exhaust gas of the semiconductor process, using a temperature sensor to measure the temperature of the exhaust gas of the semiconductor process that has passed through the VOC concentration sensor; after measuring the VOC concentration of the exhaust gas of the semiconductor process and measuring the temperature of the exhaust gas of the semiconductor process, using a VOC removal system to remove a plurality of VOCs from the exhaust gas of the semiconductor process, wherein the temperature sensor is located between the VOC concentration sensor and the VOC concentrator rotor of the VOC removal system; providing a plurality of parameters of the VOC removal system to an analysis model trained by a machine learning process, the parameters including the temperature and VOC concentration of the exhaust gas; using the analysis model to generate a predicted future VOC removal efficiency based on the parameters; using the analysis model to generate a plurality of adjustment parameters based on the predicted future VOC removal efficiency and the parameters, wherein the adjustment parameters include the recommended temperature of the VOC burner; adjusting the VOC removal system based on the adjustment parameters; and using the exhaust chimney of the VOC removal system to directly receive a first portion of the exhaust gas from the VOC concentrator rotor and directly receive a second portion of the exhaust gas from the heat exchanger of the VOC removal system. In some embodiments, the step of generating the adjustment parameters includes: using the analysis model to identify changes to the VOC removal system that are predicted to result in improved VOC removal efficiency. In some embodiments, the parameters include a current VOC removal efficiency of the VOC removal system. In some embodiments, the parameters include the rotational speed of the VOC concentrator rotor. In some embodiments, the parameters include the temperature of the VOC burner. In some embodiments, the parameters include the temperature and speed of the desorption fan. In some embodiments, the adjustment parameters include the recommended rotational speed of the VOC concentrator rotor. In some embodiments, the adjustment parameters include the recommended rotational speed of the desorption fan. In some embodiments, the adjustment parameters include the recommended desorption temperature.

[0169] In some embodiments according to the present disclosure, a volatile organic compound (VOC) removal system includes a VOC concentrator rotor, a desorption fan, a burner, a control system, a heat exchanger, and an exhaust stack. The VOC concentrator rotor has an adsorption zone, a desorption zone, and a cooling zone. The cooling zone connects the adsorption zone to the desorption zone. The adsorption zone is configured to adsorb a plurality of VOCs from a first portion of the exhaust gas from a semiconductor process, and the desorption zone is configured to desorb the VOCs to a second portion of the exhaust gas. A first temperature sensor is placed on the output side of the adsorption zone of the VOC concentrator rotor and is configured to sense a first temperature of the VOC concentrator rotor. A second temperature sensor is placed on the input side of the adsorption zone of the VOC concentrator rotor and is configured to sense a second temperature of the VOC concentrator rotor. Each of the first temperature sensor and the second temperature sensor is closer to the adsorption zone than to the desorption zone. The desorption fan is configured to drive the second portion of the exhaust gas through the desorption zone of the VOC concentrator rotor. The burner is configured to burn the VOCs from the second portion of the exhaust gas. The control system includes an analysis model. The analysis model is trained through a machine learning process and is configured to analyze a plurality of parameters of the VOC concentrator rotor, the desorption fan, and the burner, and is configured to generate a predicted future VOC removal efficiency based on the parameters, and generate a plurality of adjustment parameters based on the parameters and the predicted future VOC removal efficiency. The control system is configured to adjust the operations of the VOC concentrator rotor, the desorption fan, and the burner based on the adjustment parameters. The parameters include the temperature of the VOC concentrator rotor, and the adjustment parameters include the recommended temperature of the burner. The parameters include the first temperature sensed by the first temperature sensor and the second temperature sensed by the second temperature sensor. The heat exchanger is configured to receive heat from the second portion of the exhaust gas downstream of the burner. The exhaust stack is configured to receive directly the first portion of the exhaust gas from the VOC concentrator rotor and directly the second portion of the exhaust gas from the heat exchanger. In some embodiments, the analysis model includes an encoder and a decoder. The encoder is configured to generate a predicted future VOC removal efficiency. The decoder is configured to generate adjustment parameters. In some embodiments, the adsorption zone of the VOC concentrator rotor is located between the first temperature sensor and the second temperature sensor. In some embodiments, the heat exchanger is configured to supply heat to the second portion of the exhaust gas upstream of the desorption zone of the VOC concentrator rotor. In some embodiments, the VOC removal system further includes a VOC sensor. The VOC sensor is configured to sense the inlet VOC in the exhaust gas. In some embodiments, the VOC removal system further includes a VOC sensor. The VOC sensor is configured to sense the outlet VOC in the exhaust gas.

[0170] In some embodiments according to the present disclosure, a method for removing volatile organic compounds (VOCs) includes: measuring a first temperature of a VOC concentrator rotor using a first temperature sensor placed on an input side of an adsorption zone of the VOC concentrator rotor; transferring an exhaust gas from a semiconductor process to the VOC concentrator rotor; adsorbing a plurality of VOCs from a first portion of the exhaust gas using the VOC concentrator rotor; transferring a second portion of the exhaust gas from the VOC concentrator rotor to a heater; transferring the second portion of the exhaust gas from the heater to the VOC concentrator rotor; desorbing the VOCs from the VOC concentrator rotor to the second portion of the exhaust gas by operating a desorption fan adjacent to the VOC concentrator rotor; combusting the VOCs from the second portion of the exhaust gas using a burner; measuring a second temperature of the VOC concentrator rotor using a second temperature sensor placed on an output side of the adsorption zone of the VOC concentrator rotor, wherein a portion of the adsorption zone of the VOC concentrator rotor is located between the first temperature sensor and the second temperature sensor; generating a predicted future VOC removal efficiency at least partially based on the first temperature and the second temperature of the VOC concentrator rotor using an analysis model trained through a machine learning process; generating a plurality of adjustment parameters based on the predicted future VOC removal efficiency and the first temperature and the second temperature of the VOC concentrator rotor using the analysis model, wherein the adjustment parameters include a recommended temperature of the burner; adjusting a rotational speed of the VOC concentrator rotor and a temperature of the burner based on the adjustment parameters; and receiving the first portion of the exhaust gas directly from the VOC concentrator rotor and receiving the second portion of the exhaust gas directly from a heat exchanger using an exhaust chimney of the VOC removal system. In some embodiments, the method further includes adjusting a rotational speed of the desorption fan based on the adjustment parameters. In some embodiments, the method further includes measuring a current VOC removal efficiency and generating adjustment parameters based on the current VOC removal efficiency. In some embodiments, measuring the current VOC removal efficiency includes measuring a first VOC concentration of the exhaust gas from the semiconductor process using a first VOC concentration sensor and measuring a second VOC concentration of the exhaust gas discharged from the exhaust chimney using a second VOC concentration sensor. In some embodiments, the method further includes receiving the second portion of the exhaust gas from the burner using a heat exchanger.

[0171] The various embodiments described above can be combined to provide other embodiments. Aspects of the embodiments can be modified as necessary to incorporate the concepts of various patents, applications, and publications to provide other embodiments.

[0172] These and other changes may be made to the embodiments in light of the above detailed description. Generally, in the claims, the terms used should not be construed as limiting the claims to the specific embodiments disclosed in the specification and the claims, but should be construed to include the full scope of all possible embodiments and equivalents to which the claims are entitled. Accordingly, the claims are not limited by this disclosure.

Claims

1. A method for removing volatile organic compounds, characterized in that, comprising: using a volatile organic compound concentration sensor to measure the concentration of volatile organic compounds in an exhaust gas of a semiconductor process; after measuring the concentration of volatile organic compounds in the exhaust gas of the semiconductor process, using a temperature sensor to measure the temperature of the exhaust gas of the semiconductor process passing through the volatile organic compound concentration sensor; after measuring the concentration of volatile organic compounds in the exhaust gas of the semiconductor process and measuring the temperature of the exhaust gas of the semiconductor process, using a volatile organic compound removal system to remove a plurality of volatile organic compounds from the exhaust gas of the semiconductor process, wherein the temperature sensor is located between the volatile organic compound concentration sensor and a volatile organic compound concentrator rotor of the volatile organic compound removal system; providing a plurality of parameters of the volatile organic compound removal system to an analysis model trained by a machine learning process, the parameters including the temperature and the concentration of volatile organic compounds of the exhaust gas; using the analysis model to generate a predicted future volatile organic compound removal efficiency based on the parameters; using the analysis model to generate a plurality of adjustment parameters based on the predicted future volatile organic compound removal efficiency and the parameters, wherein the adjustment parameters include a recommended temperature of a volatile organic compound burner; adjusting the volatile organic compound removal system based on the adjustment parameters; and using an exhaust chimney of the volatile organic compound removal system to directly receive a first portion of the exhaust gas from the volatile organic compound concentrator rotor and directly receive a second portion of the exhaust gas from a heat exchanger of the volatile organic compound removal system.

2. The method according to claim 1, characterized in that, the step of generating the adjustment parameters includes: using the analysis model to identify changes to the volatile organic compound removal system that are predicted to result in improved volatile organic compound removal efficiency.

3. The method according to claim 1, characterized in that, the parameters include a current volatile organic compound removal efficiency of the volatile organic compound removal system.

4. The method according to claim 1, characterized in that, the parameters include a temperature of the volatile organic compound burner.

5. A volatile organic compound removal system, characterized in that, comprising: a volatile organic compound concentrator rotor having an adsorption zone, a desorption zone, and a cooling zone, the cooling zone connecting the adsorption zone to the desorption zone, the adsorption zone configured to adsorb a plurality of volatile organic compounds from a first portion of an exhaust gas from a semiconductor process and the desorption zone configured to desorb the volatile organic compounds into a second portion of the exhaust gas; a first temperature sensor placed on an output side of the adsorption zone of the volatile organic compound concentrator rotor and configured to sense a first temperature of the volatile organic compound concentrator rotor; A second temperature sensor disposed on the input side of the adsorption zone of the volatile organic compound concentrator rotor and configured to sense a second temperature of the volatile organic compound concentrator rotor, wherein each of the first temperature sensor and the second temperature sensor is closer to the adsorption zone than to the desorption zone; A desorption fan configured to drive the second portion of the exhaust gas through the desorption zone of the volatile organic compound concentrator rotor; A burner configured to burn the volatile organic compounds from the second portion of the exhaust gas; A control system including an analysis model that is trained by a machine learning process and is configured to analyze a plurality of parameters of the volatile organic compound concentrator rotor, the desorption fan, and the burner, and is configured to generate a predicted future volatile organic compound removal efficiency based on the parameters, and generate a plurality of adjustment parameters based on the parameters and the predicted future volatile organic compound removal efficiency, the control system being configured to adjust the operations of the volatile organic compound concentrator rotor, the desorption fan, and the burner based on the adjustment parameters, wherein the parameters include a temperature of the volatile organic compound concentrator rotor, the adjustment parameters include a recommended temperature of the burner, and the parameters include the first temperature sensed by the first temperature sensor and the second temperature sensed by the second temperature sensor; A heat exchanger configured to receive heat from the second portion of the exhaust gas downstream of the burner; and An exhaust chimney configured to receive the first portion of the exhaust gas directly from the volatile organic compound concentrator rotor and the second portion of the exhaust gas directly from the heat exchanger.

6. The volatile organic compound removal system according to claim 5, wherein, the adsorption zone of the volatile organic compound concentrator rotor is located between the first temperature sensor and the second temperature sensor.

7. The volatile organic compound removal system according to claim 5, wherein, the heat exchanger is configured to supply heat to the second portion of the exhaust gas upstream of the desorption zone of the volatile organic compound concentrator rotor.

8. A method for removing volatile organic compounds, characterized in that, comprising: measuring a first temperature of the volatile organic compound concentrator rotor by using a first temperature sensor disposed on an input side of an adsorption zone of a volatile organic compound concentrator rotor; transferring an exhaust gas from a semiconductor process to the volatile organic compound concentrator rotor; adsorbing a plurality of volatile organic compounds from a first portion of the exhaust gas by using the volatile organic compound concentrator rotor; transferring a second portion of the exhaust gas from the volatile organic compound concentrator rotor to a heater; transferring the second portion of the exhaust gas from the heater to the volatile organic compound concentrator rotor; desorbing the volatile organic compounds from the volatile organic compound concentrator rotor to the second portion of the exhaust gas by operating a desorption fan adjacent to the volatile organic compound concentrator rotor; Use a burner to burn the volatile organic compounds from the second part of the exhaust gas; Use a second temperature sensor placed on the output side of the adsorption zone of the volatile organic compound concentrator rotor to measure a second temperature of the volatile organic compound concentrator rotor, wherein a part of the adsorption zone of the volatile organic compound concentrator rotor is located between the first temperature sensor and the second temperature sensor; Use an analysis model trained through a machine learning process to generate a predicted future volatile organic compound removal efficiency based at least in part on the first temperature and the second temperature of the volatile organic compound concentrator rotor; Use the analysis model to generate a plurality of adjustment parameters based on the predicted future volatile organic compound removal efficiency and the first temperature and the second temperature of the volatile organic compound concentrator rotor, wherein the adjustment parameters include a recommended temperature of the burner; Adjust a rotational speed of the volatile organic compound concentrator rotor and a temperature of the burner based on the adjustment parameters; and Use an exhaust chimney of the volatile organic compound removal system to directly receive the first part of the exhaust gas from the volatile organic compound concentrator rotor and directly receive the second part of the exhaust gas from a heat exchanger.

9. The method according to claim 8, wherein, further comprising: adjusting a rotational speed of the desorption fan based on the adjustment parameters.

10. The method according to claim 8, wherein, further comprising: measuring a current volatile organic compound removal efficiency and generating the adjustment parameters based on the current volatile organic compound removal efficiency.

Images