Artificial Intelligence Sterilization Dryer Control Method and Sterilization Dryer Based on Thermal Load Characteristic Classification

Abstract

The present invention relates to an AI-based control technology for a sterilizing dryer that performs sterilization and drying processes in kitchens of public facilities. The present invention generates a load vector by chronologically synchronizing the rate of temperature change measured during the temperature rise section inside the sterilizing dryer chamber with the rate of humidity change measured during the humidity decrease section after reaching the target temperature. This vector is then input into a machine-learned AI model (load classification model) to precisely classify the thermal load characteristics (load class) of the object to be dried. Furthermore, the heater output and operating time are actively adjusted by applying a control profile optimized for the classified load class. In particular, the present invention effectively reduces the power consumption of the sterilizing dryer without a separate weight sensing sensor by stopping the heater operation early when drying has progressed to a certain level and finishing the drying using only the residual heat inside the chamber and the blower fan, or by performing intermittent operation when preheating is complete.

Description

Technology Field

[0001] The present invention relates to an artificial intelligence-based sterilizing dryer control technology for performing sterilization and drying processes in kitchens of public facilities. Background Technology

[0002] In kitchen facilities such as institutional cafeterias, restaurants, hospitals, and schools, sterilizing dryers are essential for the hygienic storage of tableware, cooking utensils, and various sanitary supplies after washing. Sterilizing dryers are generally equipped with heaters and fans to supply high-temperature hot air into the chamber, thereby removing residual moisture from the items to be dried and inhibiting bacterial growth.

[0003] Conventional sterilization dryers simply operated a heater for a user-set duration; however, to save energy, a method utilizing temperature and humidity sensors to stop the heater when the internal humidity reaches a certain level has been proposed. Nevertheless, this control method also had limitations, as it relied on controlling the system based on whether the current humidity value fell below a pre-set threshold. This is because a simple comparison of sensor values ​​makes it impossible to distinguish the characteristics of the thermal load, such as whether humidity is dropping slowly because the chamber is full of items to be dried, or drying is rapid because there are almost no items.

[0004] Ultimately, existing control methods had an inefficient technical limitation in that they relied solely on comparing threshold values ​​of set times or humidity levels, regardless of the thermal load characteristics of the material being dried, resulting in the input of excessive energy. Since this leads to unnecessary energy consumption, there is an urgent need to introduce intelligent control technology that actively regulates the operation of sterilization dryers to achieve carbon neutrality and energy conservation. Prior art literature

[0005] Korean Intellectual Property Office Registered Patent Publication No. 10-1994840 "High-efficiency sterilizing dryer and method for controlling operation thereof" The problem to be solved

[0006] The present invention aims to classify thermal load characteristics, including the amount and drying state of the material to be dried, by analyzing the temperature and humidity change patterns inside a sterilization dryer chamber, and to actively control a heater according to the thermal load characteristics of the material to be dried. Through this, the present invention aims to prevent the input of unnecessary heat into the chamber by controlling the output of the heater and to reduce the energy required for the sterilization drying process. means of solving the problem

[0007] The artificial intelligence sterilization dryer control method based on thermal load characteristic classification according to the present invention is a method performed in a processor controlling a sterilization dryer, comprising: a step of generating temperature data within a chamber; a step of generating humidity data within the chamber; a step of generating a load vector including a rate of change in temperature and a rate of change in humidity per unit time using the temperature data and the humidity data; a step of inputting the load vector into a load classification model, which is an artificial intelligence model, to classify a load class for the thermal load characteristics of an object to be dried within the chamber or the operating state of the sterilization dryer; and a step of controlling a heater according to a control profile corresponding to the load class, wherein the steps of generating the load vector and controlling the heater are repeated at a predetermined period.

[0009] The step of generating the load vector involves calculating the rate of change in temperature within the chamber during a first interval until the temperature within the chamber reaches a predetermined target temperature, and the rate of change in humidity within the chamber during a second interval from when the temperature within the chamber reaches the target temperature until sterilization drying is completed, and combining the rates of change at different points in time into a single vector.

[0011] The step of generating the load vector comprises: a step of calculating the rate of change in temperature in the first section; and a step of calculating the rate of change in humidity at the predetermined intervals in the second section and updating the load vector by combining it with the rate of change in temperature, and the step of classifying the load class comprises inputting the updated load vector into the load classification model at the predetermined intervals.

[0013] The step of generating the load vector above is calculated by dividing the rate of temperature change per unit time, based on the time required to reach the target temperature, by the reference rate of temperature change when the inside of the chamber is empty.

[0015] The above load classification model is a machine learning model trained by supervised learning using training data that includes input values ​​including the rate of change in temperature and the rate of change in humidity collected in the first and second sections, and output values ​​of load classes classified based on the actual amount of material to be dried and the drying state under the same conditions.

[0017] The load classification model described above is a model learned through unsupervised learning based on accumulated distribution data of the load vectors, and classifies each cluster distinguished based on similarity between the load vectors in the feature space of the load vectors formed by the temperature change rate and humidity change rate as a load class.

[0019] The above load classification model is a model learned through reinforcement learning, defining the load vector as a state and the selection of the load class as a behavior, and its parameters are updated in a direction that maximizes the reward calculated based on the drying success rate according to whether the target humidity is reached and the energy efficiency according to the amount of power input, as a result of performing a control profile according to the selected load class.

[0021] The step of controlling the heater involves, when the load class is classified into a first class distributed in an area having a relatively high temperature change rate and a relatively low humidity change rate in the feature space of the load classification model, setting the output ratio of the heater lower than a preset reference output ratio and setting the operating time of the heater shorter than a preset time; when the load class is classified into a third class distributed in an area having a relatively low temperature change rate and a relatively low humidity change rate in the feature space, setting the operating time of the heater shorter than a preset time; and when the load class is classified into a fourth class distributed in an area having a relatively low temperature change rate and a relatively high humidity change rate in the feature space, setting the output ratio of the heater higher than a preset reference output ratio or to a maximum value.

[0023] The step of classifying the load class is to classify it into a fifth class regardless of the humidity change rate when the temperature change rate is less than a predetermined threshold, and the step of controlling the heater is to set the operating time of the heater to a restart time shorter than a preset time when classified into the fifth class, and to perform intermittent operation by operating the heater again for the restart time after a preset waiting time following the shutdown of the heater.

[0025] An artificial intelligence sterilizing dryer based on thermal load characteristic classification according to the present invention comprises: a sensor unit for detecting temperature and humidity within a chamber; a heating unit including a heater for heating the chamber; and a control unit including a memory for storing commands and a processor for executing commands loaded from the memory. The commands enable the execution of a control method for an artificial intelligence sterilizing dryer based on thermal load characteristic classification. Effects of the invention

[0026] The present invention analyzes a load vector combining the rate of change in temperature and the rate of change in humidity inside a sterilization dryer chamber using a load classification model, thereby enabling precise classification of the thermal load characteristics (heat capacity and drying state, etc.) of the object to be dried without a separate weight sensing sensor. Accordingly, since heater control optimized for the actual load characteristics of the object to be dried is possible, unnecessary power waste caused by uniform control of the heater can be prevented, thereby enabling efficient energy saving.

[0027] Furthermore, the present invention maximizes energy efficiency by applying differentiated control profiles according to classified load classes. In particular, when the heat capacity is high but drying is in progress, the heater power is cut off early and residual heat drying is performed using the thermal inertia of the material to be dried; and when preheating is complete and continuous operation is in progress, short-cycle intermittent operation is performed, thereby preventing overheating and minimizing the heater operating time.

[0028] Furthermore, the present invention updates the parameters of an artificial intelligence model to maximize power efficiency based on the correlation between drying completion and power consumption through various machine learning methods. This enables active and sustainable energy management, in which the load classification model independently finds the optimal energy efficiency point, even if external variables such as the exhaust environment at the installation site or equipment aging change.

[0029] Meanwhile, even if an effect is not directly described herein, any effect expected or anticipated by the description of the invention should be understood as being described in the effects of the invention. Brief explanation of the drawing

[0030] FIG. 1 is a drawing showing a sterilizing dryer in which an artificial intelligence sterilizing dryer control method based on thermal load characteristic classification according to an embodiment of the present invention is performed. FIG. 2 is a diagram showing the configuration of a sterilization dryer control method according to an embodiment of the present invention. FIG. 3 is a diagram showing the process of generating a load vector in a sterilization dryer control method according to an embodiment of the present invention. FIG. 4 is a diagram showing the process of calculating the rate of change of temperature among load vectors in a sterilization dryer control method according to an embodiment of the present invention. FIG. 5 is a diagram showing the distribution of load vectors in a sterilization dryer control method according to an embodiment of the present invention. FIG. 6 is a diagram showing the configuration of a load classification model in a sterilization dryer control method according to an embodiment of the present invention. FIG. 7 is a diagram showing the heater control process by load class in the sterilization dryer control method according to an embodiment of the present invention. FIG. 8 is a diagram showing the heater control process according to the 5th class in the sterilization dryer control method according to an embodiment of the present invention. FIG. 9 is a diagram showing the configuration of a sterilization dryer according to an embodiment of the present invention. Specific details for implementing the invention

[0031] The present invention is not limited to the embodiments described in the description of the invention and can be implemented in various ways. The terms used in the description of the invention are intended to describe the embodiments and are not intended to limit the invention. In the description of the invention, the term "comprising" is intended to indicate the existence of a configuration in which the described features are combined. Therefore, it should be understood that this does not exclude the existence or addition of one or more other features or configurations combining them. Detailed descriptions of configurations and functions known to those skilled in the art to which the present invention pertains are omitted below.

[0033] FIG. 1 is a drawing showing a sterilizing dryer (100) in which an artificial intelligence sterilizing dryer control method based on thermal load characteristic classification according to an embodiment of the present invention is performed. FIG. 9 is a drawing showing the configuration of a sterilizing dryer (100) according to an embodiment of the present invention. Referring to FIG. 1, the sterilizing dryer (100) includes a chamber that forms a space inside that is isolated from the outside. Various items to be dried, including sanitary products used in a kitchen, are loaded in the chamber, and the sterilizing dryer (100) performs a sterilization and drying process of the items to be dried by circulating high-temperature hot air inside the sealed chamber. Referring to FIG. 1 and FIG. 9, the sterilizing dryer (100) to which the artificial intelligence sterilizing dryer control method based on thermal load characteristic classification according to an embodiment of the present invention (hereinafter referred to as the 'sterilizing dryer control method') is applied includes a sensor unit (110), a heating unit (120), a ventilation unit (130), and a control unit (140). According to an embodiment, the sterilizing dryer (100) may further include a sterilization unit (150).

[0034] The sensor unit (110) detects the physical state inside the chamber in real time. The sensor unit (110) may include a temperature sensor that measures the temperature inside the chamber and a humidity sensor that measures the humidity. The sensor unit (110) detects the temperature and humidity in real time so that the control unit (140) can calculate the rate of change per unit time. The heating unit (120) provides a heat source that heats the air inside the chamber to evaporate the moisture of the object to be dried and sterilizes the object to be dried. The heating unit (120) includes a heater that converts electrical energy into thermal energy, and the operating time or output ratio can be precisely controlled according to the control signal of the control unit (140). For convenience of explanation in this specification, the heating unit (120) may be described by replacing it with a heater, which is one of the embodiments. The ventilation unit (130) forcibly circulates the air inside the chamber and discharges the moisture evaporated from the object to be dried to the outside. The ventilation unit (130) may include a blower fan or blower that rotates by a motor. In particular, the ventilation unit (130) can be controlled independently regardless of whether the heating unit (120) is operating, thereby enabling a residual heat drying process that circulates residual heat (residual heat) remaining in the chamber to the object to be dried. The sterilization unit (150) is configured to kill bacteria in the object to be dried and may include an ultraviolet (UV) lamp or a plasma generation module. However, according to the embodiment, the present invention may be configured to perform a sterilization function using only high-temperature hot air from the heating unit (120) without the sterilization unit (150). The control unit (140) includes a processor that controls the sterilization dryer (100) and a memory in which commands implementing the control method are stored. The control unit (140) analyzes temperature and humidity data collected from the sensor unit (110) and inputs this into a pre-trained artificial intelligence model (load classification model) to classify the thermal load characteristics of the object to be dried.And by actively variably controlling the output of the heater according to the optimal profile corresponding to the classified load characteristics, energy consumption is reduced while maintaining the drying efficiency of the sterilization dryer (100).

[0036] The following describes specific steps for a sterilization dryer control method. FIG. 2 is a diagram showing the configuration of a sterilization dryer control method according to an embodiment of the present invention. Referring to FIG. 2, the sterilization dryer control method according to an embodiment of the present invention includes a temperature data generation step (S100), a humidity data generation step (S200), a load vector generation step (S300), a load class classification step (S400), and a heater control step (S500). Each step is performed by a processor of a control unit (140) embedded in the sterilization dryer (100). For convenience of explanation, the processor performing each step may be omitted in the following description.

[0037] The temperature data generation step (S100) and the humidity data generation step (S200) are processes for collecting physical environmental information inside the chamber. When the sterilization dryer (100) starts operating, the sensor unit (110) measures the temperature and humidity inside the chamber in real time, and the control unit (140) generates and stores the temperature and humidity detected by the sensor unit (110) as time-series data including time information. The data generated in these steps is utilized as data for generating a load vector, which will be described later, beyond simple monitoring of the current state.

[0038] The load vector generation step (S300) is a process of processing the generated temperature and humidity data into data that can be recognized by an artificial intelligence model. The control unit (140) generates a 'rate of change in temperature per unit time' and a 'rate of change in humidity per unit time' by differentiating or performing regression analysis on the accumulated temperature and humidity data. Then, using the generated rates of change, it generates a load vector representing coordinates in a multidimensional space. Here, the load vector serves as an indicator that implicitly represents the load characteristics of the object to be dried (load amount and residual moisture, etc.) and the current drying progress (amount of moisture evaporation).

[0039] The load class classification step (S400) is a process of classifying the thermal load characteristics of the object to be dried or the operating state of the sterilization dryer based on the generated load vector. The control unit (140) inputs the generated load vector as an input value to a load classification model, which is a pre-trained artificial intelligence model. The load classification model classifies the characteristics of the input load vector based on knowledge learned through machine learning, and as a result, outputs a load class corresponding to the thermal load characteristics of the object to be dried. For example, the load class can be classified into multiple classes depending on whether the amount of the object to be dried is large or small, or whether the drying state is wet or dry.

[0040] The heater control step (S500) is a process of applying customized control according to the classified load class. The control unit (140) loads a control profile corresponding to the classified load class. The control profile defines the target temperature, operating time, and output ratio of the heater that is most efficient for the corresponding load characteristics. By actively controlling the heater according to this profile, the control unit (140) prevents excessive heating and maximizes the use of residual heat to minimize the power consumption of the sterilization dryer.

[0041] In particular, the sterilization dryer control method according to an embodiment of the present invention involves the load vector generation step (S300) to the heater control step (S500) being repeatedly performed at predetermined intervals. This is because as the sterilization drying process progresses, the moisture content of the material to be dried continuously decreases, and accordingly, the thermal characteristics (load vector) inside the chamber also change dynamically. For example, even if the material to be dried was initially classified as a 'high load' class due to high moisture content, it may be reclassified as a 'low load' class as drying progresses. Accordingly, the control unit (140) monitors the periodically updated load class to dynamically lower the output of the heater or predicts the drying completion time and performs real-time feedback control to stop the heater operation early. Through this, the present invention can effectively reduce energy waste compared to conventional sterilization dryer control methods that control the heater in a fixed pattern.

[0043] FIG. 3 is a diagram illustrating the process of generating a load vector in a sterilization dryer control method according to an embodiment of the present invention. Referring to FIG. 3, the control unit (140) divides the entire sterilization drying process into a first section and a second section, and performs a data processing process of extracting the rate of change of different physical quantities from each section and combining them into a single load vector.

[0044] First, the first section is the time when the heater started operating (t start From ) the internal chamber temperature to the preset target temperature (T targetIt is defined as a heating section up to the point (t1) at which the temperature is reached. In this section, the control unit (140) calculates the rate of temperature change (Δt) corresponding to the temperature increase per unit time. Thermodynamically, when the same amount of heat is supplied, the rate at which the temperature rises is inversely proportional to the total heat capacity of the object to be dried, that is, the mass of the loaded object. Therefore, the rate of temperature change (Δt) measured in the first section serves as an indicator representing the total load of the object to be dried. The control unit (140) stores the rate of temperature change (ΔT) in memory at the point (t1) when the first section ends. The unit time for calculating the rate of temperature change (ΔT) in the first section can be set in various ways. This value remains a constant during the drying process unless the input amount of the object to be dried is changed midway. Subsequently, the second section is defined from the point (t1) when the internal chamber temperature reaches the target temperature until the point (t) when sterilization drying is completed. end It is defined as a drying section up to ). In this section, the control unit (140) calculates a humidity change rate (ΔH) corresponding to the humidity drop rate per unit time at predetermined intervals. Unlike the temperature change rate of the first section, the humidity change rate corresponds to a dynamic value that changes moment by moment from a high change rate at the beginning of drying to a low change rate at the end of drying as the moisture of the object to be dried evaporates. Therefore, the control unit (140) can repeatedly calculate the temperature change rate.

[0045] The main feature of the present invention is the first section (t start ~ t1) and the second interval (t1~ t endThe feature is that data from two different points in time is synchronized to form a single input vector. Specifically, the control unit (140) calculates only the temperature change rate (ΔT) while excluding the humidity change rate in the first section, and in the second section, combines the temperature change rate (ΔT) of a past point in time (before t1) stored in memory with the humidity change rate (ΔH) of the current point in time to update a new load vector (ΔT, ΔH) every cycle. This corresponds to a feature of the present invention that is clearly distinguished from conventional control technology that simply collects and controls temperature data and humidity data at the same point in time. Since the first section is a section where the internal temperature of the chamber rises rapidly due to heater heating, humidity is prone to drop in inverse proportion to the rise in temperature regardless of the amount of moisture evaporation, or noise is likely to occur due to irregular turbulence. That is, since the first section corresponds to a heating section before moisture evaporation begins in earnest, the humidity data at this point in time does not reflect the actual wet state of the object to be dried. Therefore, the present invention improves the accuracy of determining the rate of change in humidity by selectively classifying and calculating only the humidity data from the second section, where evaporation is active. Furthermore, if the rate of change in temperature is measured in the second section, the temperature converges to zero, and unlike the heating section, there is insufficient data to infer the thermal load characteristics of the object to be dried, resulting in a failure to properly reflect the characteristics of the object. Accordingly, the present invention maintains past values ​​for the rate of change in temperature as a measure for judging the magnitude of the load, and reflects the current value for the rate of change in humidity as a measure for judging the progress of drying, thereby enabling the three-dimensional and accurate modeling of the load characteristics of the object to be dried. Additionally, the reason for synchronizing data from different time points into a single vector in the present invention is to simultaneously determine the total mass and drying state of the object to be dried without the addition of separate sensors. The rate of change in temperature (ΔT) has the nature of a constant representing the total heat capacity of the object to be dried, while the rate of change in humidity (ΔH) has the nature of a variable representing how much moisture remains.If control is performed solely based on the rate of change in humidity without combining the two characteristics, a problem arises in that it is impossible to distinguish whether the humidity is dropping quickly because the load is small or because drying is already complete. Therefore, the present invention combines the previously measured rate of change in temperature (a measure of load size) with the current rate of change in humidity (a measure of progress rate) to enable three-dimensional and precise inference of the state inside the chamber based on the classification of thermal load characteristics, such as "the [large volume] of material to be dried has entered the [final drying stage]." Through this real-time vector update mechanism, the control unit (140) recognizes the state where the drying of the material to be dried is not completed (ΔH1) immediately after the drying process starts and operates the heater strongly, but at the point where the rate of change in humidity becomes gradual (ΔH2), the load vector is updated so that the drying can be reclassified as having progressed beyond a certain level. Consequently, the present invention performs adaptive control by recalculating the optimal load class and controlling the heater at every point in time according to the changing thermal load state inside the chamber, rather than a fixed scenario.

[0047] FIG. 4 is a diagram illustrating the process of calculating the rate of temperature change among load vectors in a sterilization dryer control method according to an embodiment of the present invention. Even if the rate of temperature change is the same, it may signify a 'high load state' in a small sterilization dryer, whereas it may signify a 'low load' in a large sterilization dryer. In this case, inefficiency occurs in that thermal load characteristics must be classified by considering the characteristics of each sterilization dryer model. Accordingly, the present invention performs a correction process of dividing the measured rate of temperature change by a reference rate of change in order to classify consistent thermal load characteristics even in sterilization dryers with different volumes. That is, the present invention performs a data preprocessing process to standardize thermal load characteristics by eliminating deviations according to the hardware specifications of the sterilization dryer (such as the volume of the chamber and the capacity of the heater).

[0048] Specifically, the control unit (140) calculates the rate of temperature change per unit time by dividing it by the reference rate of temperature change when the chamber is empty. That is, instead of generating the rate of temperature change per unit time of the temperature data as is as a load vector, it performs a process of correcting the rate of temperature change. Referring to FIG. 4, the graph indicated by the dotted line is the temperature rise curve when the chamber is empty, and the graph indicated by the solid line is the actual temperature rise curve when the material to be dried is loaded. At this time, the end point of the first section, i.e., the target temperature (T target The time taken to reach ) depends on the amount of material to be dried (load). Due to the low heat capacity of the empty chamber, a short time (t ref While it reaches the target temperature in just ), the chamber loaded with the material to be dried takes a longer time (t) due to its high thermal capacity. act ) takes (t act > t ref ). Therefore, the control unit (140) has the same temperature range (T start ~ T target The average slope within is calculated and compared. That is, the rate of temperature change is calculated based on the time required to reach the target temperature.

[0049] The control unit (140) operates the heater in an empty chamber state when the sterilization dryer is shipped or in the initial setting mode, and the time (t) taken to reach the target temperature from the starting temperature. ref Measure ) and the resulting reference temperature change rate per unit time (S ref ) is stored in memory as a reference value. And during actual drying operation, the control unit (140) stores the time (t) required to reach the same target temperature. act Measure ) and the resulting actual rate of temperature change per unit time (S act ) calculates. Then, the control unit (140) calculates the normalized index (L) according to the following mathematical formula 1. norm It calculates ).

[0050] [Mathematical Formula 1]

[0051] L norm = S act / S ref

[0052] The control unit (140) uses a normalization index as the rate of change of temperature of the load vector. Here, the normalization index has a value between 0 and 1, and a value closer to 1 indicates 'low load', and a value closer to 0 indicates 'high load'. Meanwhile, the present invention sets an effective calculation interval to prevent the problem of calculation errors occurring when the slope converges to 0 at the point of reaching the target temperature. The control unit (140) may limit the effective calculation interval to the point where the temperature reaches a certain percentage (e.g., 90%) of the target temperature or is lower than the target temperature by a predetermined margin value (e.g., 5°C). That is, the control unit (140) secures the reliability of temperature data for classifying thermal load characteristics by calculating the rate of change of temperature by extracting only data from the pure heating interval before the slope is distorted by the intervention of heater control. In addition, if the initial temperature at the start of operation of the sterilization dryer is already close to the target temperature (continuous operation) or is too high to secure an effective calculation interval, the control unit (140) omits the calculation of the temperature change rate and classifies the state as preheating complete to perform customized control.

[0054] FIG. 5 is a diagram showing the distribution of load vectors in a sterilization dryer control method according to an embodiment of the present invention. Referring to FIG. 5, the load vector generated by the control unit (140) can be projected and mapped onto a two-dimensional vector space in which the rate of change of temperature (ΔT) is the first axis (X-axis) and the rate of change of humidity (ΔH) is the second axis (Y-axis). In this space, the state of the object to be dried forms a unique distribution in different regions according to physical characteristics. For example, Class 1 is a region where the rate of change of temperature (ΔT) is high and the rate of change of humidity (ΔH) is low. This may mean a state where the amount of the object to be dried is small, such as a small amount of dry food tray, so the temperature rises quickly, and the amount of moisture evaporation is low because it is already dried to a certain degree. Class 4 is a region where the rate of change of temperature (ΔT) is low and the rate of change of humidity (ΔH) is high. This may mean a state where the amount of the object to be dried is large, such as a large amount of wet food tray, so the temperature rises slowly, and active evaporation occurs due to the high moisture content. In this way, the present invention enables an artificial intelligence model to classify the fine thermal load characteristics of an object to be dried by combining two time-series separated physical quantities into a load vector.

[0055] Additionally, the load classification model can classify error classes when the operation deviates from the normal operating range. Error classes include conditions where the heater operates normally but air circulation stops due to a failure of the blower fan, or where localized overheating is detected because the temperature sensor is blocked by the material being dried. These error classes exhibit characteristics where the rate of change in temperature (ΔT) is abnormally high and the rate of change in humidity (ΔH) is very low due to the absence of air circulation. However, in the characteristic space formed by the load vector, the error class may be very close to the first class or partially overlap with it. Since both the empty chamber state and the blower fan failure exhibit rapid temperature rise and low humidity change, it is difficult to clearly distinguish between the two using conventional control logic that judges based simply on a threshold value. If the control unit (140) mistakes an error situation for the first class, the heater will continue to operate in the failure state, posing a risk of fire or damage to the sterilizing dryer. Furthermore, there are overlapping areas between classes corresponding to the normal operating range due to interference in physical characteristics. For example, Class 2, in which stainless steel plates with high thermal conductivity are loaded in a standard amount, and Class 3, in which ceramic or plastic tableware with high heat capacity is loaded in large quantities but drying has progressed beyond a certain level, may exhibit similar rates of temperature change (ΔT) and humidity change (ΔH) in specific sections despite differences in actual loading amount and material. In other words, in a real environment where various variables such as the material of the object to be dried and the loading method interact in a complex manner, classes with different load characteristics are not clearly distinguished linearly in the feature space of the load vector and are distributed forming irregular boundaries.

[0056] Therefore, the present invention utilizes a load classification model, which is an artificial intelligence model capable of learning the non-linear characteristics of data, to precisely determine each class based on subtle differences in physical characteristics. Through machine learning, the load classification model can accurately classify load classes by detecting even minute differences in load vectors that are difficult for humans or conditional control logic to identify.

[0058] FIG. 6 is a diagram showing the configuration of a load classification model in a sterilization dryer control method according to an embodiment of the present invention. The load classification model of the present invention can be learned through supervised learning, unsupervised learning, and / or reinforcement learning methods depending on the purpose of application and the data acquisition environment.

[0059] In a supervised learning method, the load classification model can be trained using training data that includes input values, such as the rate of change in temperature and the rate of change in humidity collected in the first and second sections, and output values, such as load classes classified based on the actual amount of material to be dried and the drying state under the same conditions. Referring to Fig. 6, in a supervised learning method, the load classification model is trained using training data composed of pairs of input values, such as load vectors, and corresponding output values, such as load classes. The load classification model corresponds to a multi-class classification model that outputs probabilities for each class. The load classification model may have an artificial neural network structure, and a softmax function may be used as the activation function of the output layer to output the probability of belonging to each class by non-linearly transforming the input feature values. The softmax function normalizes the model's output values ​​to probability values ​​between 0 and 1 to output the probability corresponding to each class. However, depending on the design method of the load classification model, various activation functions such as sigmoid, hyperbolic tangent (Tanh), or ReLU (Rectified Linear Unit) may be applied. Training data for training a load classification model can be generated through iterative experiments by introducing quantified materials to be dried (e.g., 10, 50, and 100 food trays, etc.) into a chamber and setting different drying states (wet, semi-dry, and dry, etc.). The collected rates of temperature change and humidity change serve as input values, and the classification class based on the actual load amount serves as the output value. The load classification model analyzes the correlation between the input change rate vector and the correct class and updates weights in a direction that minimizes the error between the predicted value and the correct value. The load classification model may use a Cross-Entropy Loss function as the loss function to calculate the difference (error) between the predicted value and the actual correct value for the input load vector.However, loss functions are not limited to this, and various loss functions such as Mean Squared Error (MSE), Hinge Loss, or Focal Loss to address data imbalance issues can be used. Load classification models iteratively update weights and biases through optimization algorithms, such as gradient descent, in a direction that minimizes the result of the loss function. This supervised learning method can establish clear classification criteria in structured environments, thereby guaranteeing high initial classification accuracy.

[0060] Next, the load classification model can be trained through unsupervised learning based on accumulated distribution data of load vectors. Unsupervised learning is a learning method for self-optimizing classification criteria by reflecting the environmental characteristics of the site where the actual sterilizing dryer (100) is installed. Since supervised learning is based on data with correct answers, the initial performance is excellent, but there is a risk that classification accuracy may decrease if the installation environment (e.g., length of exhaust duct, humidity of the kitchen, and seasonal temperature differences, etc.) differs from the factory shipment standards. To compensate for this, the load classification model of the present invention can accumulate data on the rate of change of temperature and the rate of change of humidity inside the chamber as the operation of the sterilizing dryer (100) accumulates, and classify load classes by self-defining them through analysis of their distribution characteristics. The control unit (140) continuously maps the load vectors calculated whenever the sterilizing dryer (100) is operated onto a feature space. When data is sufficiently accumulated, the control unit (140) performs a clustering algorithm such as K-means clustering or DBSCAN (Density-Based Spatial Clustering of Applications with Noise). At this time, the clustering algorithm calculates the Euclidean distance or cosine similarity between load vectors in a vector space to group data that are close to each other and have high density to form a single cluster. For example, if the most frequently occurring operating patterns in a specific kitchen where a sterilizing dryer (100) is installed form a large cluster, the control unit (140) defines this as a standard class. Conversely, if data is generated that deviates by more than a preset threshold distance from the center of the standard class, it is classified into respective load classes considering load characteristics and a control profile is applied accordingly.By applying this unsupervised learning method, the present invention can learn the unique physical environment of individual sites where the sterilizing dryer (100) is installed without a separate data labeling process, and actively establish precise load classification criteria optimized for the site over time. The load classification model based on unsupervised learning has the advantage of being able to automatically reflect changes in heater performance or sensor sensitivity that change due to equipment aging into the classification criteria.

[0061] The load classification model can be learned through reinforcement learning by defining the load vector as the state and the selection of the load class as the action. The reinforcement learning method can be applied to intensively solve the optimization problem of maximizing energy efficiency, beyond the control objective of the sterilization dryer (100) to complete drying. When the load classification model is learned through reinforcement learning, the control unit (140) is defined as an agent interacting with the chamber, which is the environment. At this time, the agent is in a state (S) at the current time point (t). t Observe ) and the corresponding action (A t Selects ), and as a result, the reward (R t It learns by repeating the process of acquiring ). Specifically, the state (S t ) is defined as a load vector that is repeatedly updated. That is, the temperature rising trend and humidity falling trend inside the current chamber become the environmental state perceived by the agent. And the behavior (A t) is defined as the selection of a load class output by the load classification model. Here, the load classification model learns a policy that predicts which load class (behavior) should be selected from the current load vector (state) to maximize the final reward, going beyond simply matching the pattern of the input data. For example, the decision itself—such as the control unit (140) determining the current state as a fourth class (wet / overloaded) and selecting a strong heater output, or determining it as a second class (standard) and selecting an intermediate output—becomes the optimal action to maximize the reward. In the reinforcement learning process, the agent must simultaneously satisfy two conflicting goals for controlling the sterilization dryer: the first goal (hygiene) of drying the object to be dried and the second goal (cost minimization) of using the minimum amount of power. Therefore, the reward function according to the embodiment of the present invention can be defined as a linear combination of a positive (+) reward (reward) based on the success or failure of drying and a negative (-) reward (penalty) based on the amount of power consumed, as shown in Equation 2 below.

[0062] [Mathematical Formula 2]

[0063] R total = α × I(H final ≤ H target ) - β × P consumption

[0064] Here, R total is the final reward value, H final ε is the final humidity inside the chamber at the time of control termination, H target is the target humidity (e.g., 5%), I is an indicator function that returns 1 if the condition is met, otherwise P consumptionrepresents the accumulated total power consumption (kWh). α and β are weights assigned to the drying success rate and energy savings, respectively. In this case, the total power consumption is reflected as the result of controlling the heater by applying a control profile based on the repeatedly updated load class. The fact that the load classification model learns to maximize rewards means that it finds the most accurate and efficient class. For example, if the agent misclassifies (behaves) it as 'high load' even though it is actually almost dry (low load), the power consumption (P) due to unnecessary heater operation... consumption ) increases rapidly, resulting in a large penalty (negative compensation). Conversely, if it is misclassified as 'low load' when it is actually wet (high load), the heater turns off prematurely, causing the target humidity (H target Since ) is not achieved, the reward (α) is not obtained. The control unit (140) updates parameters to accurately select the only class capable of achieving a balance point between drying success and energy saving in a given load vector (state) through Q-learning or a policy gradient algorithm. The control unit (140) performs Q-learning or a policy gradient algorithm, etc., based on a reward system. In the early stages of learning, the agent may randomly select a class and experience drying failure (insufficient humidity) or consume excessive power, but as trial and error accumulates, the final reward value (R total The parameters are updated in a direction that maximizes ). As a result, the load classification model, having completed reinforcement learning, establishes on its own an optimal policy to suppress heater output as much as possible up to a lower limit where the target humidity can be reached. Through this, the present invention implements customized control that simultaneously achieves two objectives—guaranteeing drying quality and maximizing energy savings—which are difficult to achieve with fixed rule-based control.

[0066] FIG. 7 is a diagram illustrating the heater control process by load class in a sterilization dryer control method according to an embodiment of the present invention. The control unit (140) controls the heater by mapping an optimal control profile according to the load class classified through a load classification model. Through this, the present invention can perform customized control according to the state of the object to be dried. For convenience of explanation, the following description is based on a food tray as an example of the object to be dried. However, the object to be dried may include various sanitary products in addition to food trays.

[0067] Referring to FIGS. 5 and 7, the second class is a reference load representing the standard loading amount and drying state of a sterilizing dryer, forming a central cluster in the distribution on the feature space of the load vector, and can serve as a criterion for relative comparison to distinguish other classes.

[0068] Class 1 is a region distributed in a feature space where the rate of change of temperature (ΔT) is relatively high and the rate of change of humidity (ΔH) is relatively low, and corresponds to cases where the amount of food trays input is very small or a small amount of food trays in a dry state that has already been dried to some extent are input. In Class 1, the control unit (140) sets the output ratio of the heater to a level lower than the preset output ratio (e.g., 50%) and also sets the operating time to be shorter than the preset time. This is because Class 1 can quickly reach the target temperature with only low output since the heat capacity of the material to be dried is small, and since there is almost no moisture to remove, there is no need to heat for a long time. This prevents unnecessary power waste caused by an excessive rise in temperature.

[0069] Class 3 is distributed in a region where both the rate of change in temperature (ΔT) and the rate of change in humidity (ΔH) are relatively low in the characteristic space. This may apply to cases where a large number of food trays are loaded, but some moisture has already been removed after washing or drying has progressed beyond a certain level. At this time, because the total mass of the loaded food trays is large, a certain level of heater operation is required to raise the internal temperature of the chamber to the target temperature. However, since the amount of moisture remaining on the surface of the food trays is small and the required amount of latent heat of evaporation consumed for moisture vaporization is low, continuously operating the heater beyond a certain level after reaching the target temperature constitutes a waste of energy. The control unit (140) maintains the output ratio of the heater at a preset reference output ratio and sets the operating time of the heater to be shorter than the preset time. Here, the reference output ratio may be the output ratio set in the conventional control method, but it can be set in various ways depending on the characteristics of the sterilization dryer (100). In particular, the control unit (140) utilizes the heat retention of the large number of loaded food trays to actively intervene in a residual heat drying state by shutting off the heater early and driving only the blower fan. In other words, energy efficiency is maximized through a residual heat drying section that dries residual moisture using only the heat emitted by the material to be dried.

[0070] Class 4 is distributed in a region where the rate of change of temperature (ΔT) is relatively low and the rate of change of humidity (ΔH) is relatively high in the characteristic space. This may apply to cases where a large number of food trays are loaded and a large amount of moisture remains on their surface. Class 4 corresponds to the load conditions that require the most energy in the drying process. In this case, not only is the total mass of the material to be dried large and thus the heat capacity high, but a large amount of latent heat of evaporation is continuously consumed to change the moisture of the food trays into a gaseous state. If the heater output is lowered for the purpose of saving energy in such a situation, the internal temperature of the chamber is actually lost due to the heat of vaporization, causing the drying time to increase exponentially, or the process may end before drying is completed, leading to hygiene problems such as the proliferation of bacteria. Therefore, the control unit (140) supplies a powerful hot air into the chamber by setting the output ratio of the heater higher than a preset reference output ratio or to the maximum allowable value of the hardware. Through this, the control unit (140) evaporates the moisture of the object to be dried to below a critical point in the shortest possible time, and thereafter, when the rate of change in humidity decreases and the classification state changes from the 4th class to another class (e.g., 3rd class or 2nd class), the heater output is controlled to save energy. In this way, the present invention does not pursue only uniform energy saving, but determines when energy should be used and when it should be saved according to the thermal load characteristics diagnosed by the load classification model, and performs optimal control according to the situation inside the chamber, thereby satisfying both drying performance (hygiene) and energy efficiency (cost).

[0072] FIG. 8 is a diagram illustrating the heater control process according to the fifth class in a sterilization dryer control method according to an embodiment of the present invention. Referring to FIG. 5 and FIG. 8, the control unit (140) classifies the load class as the fifth class regardless of the humidity change rate (ΔH) when the temperature change rate (ΔT) converges to zero or is below a preset threshold value during the load class classification step. This may correspond to a continuous operation situation where the sterilization dryer is operated again immediately after completing a previous operation. In this case, since the inside of the chamber and the object to be dried (food tray, etc.) are already heated above a certain temperature, the temperature does not rise rapidly even if the heater is operated. If, in this situation, it is mistakenly identified as the fourth class simply because the temperature change rate is low, the maximum heat source is introduced into the already hot chamber, resulting in severe overheating and unnecessary power waste. Therefore, the control unit (140) performs intermittent operation during the heater control step when it is classified as the fifth class. Specifically, the control unit (140) has a restart time (t) shorter than the operating time of the heater than the preset time. on Set to ), and after stopping the heater, a predetermined waiting time (t off After waiting for ) and then restarting the heater again for the above restart time (t on Intermittent control is performed, operating only during ). At this time, restart time (t onThe time is set to replenish a minimum amount of heat sufficient to prevent the temperature inside the chamber from dropping below the effective sterilization range. That is, the control unit (140) controls the heater in a way that compensates only for the amount of heat lost to the outside through the chamber walls, etc., rather than the energy used to heat the object to be dried from the beginning. The control unit (140) sets the output ratio of the heater to a minimum output ratio (50% in the example of FIG. 8) that is lower than a preset standard output ratio and can compensate for the amount of heat lost. As a result, in a mass catering facility environment where continuous use is frequent, the present invention maximizes user convenience and energy efficiency simultaneously by performing maintenance operation that minimizes energy consumption by automatically detecting the continuous operation state through a load classification model without the user having to set it separately.

[0074] Referring to FIG. 9, an artificial intelligence sterilizing dryer based on thermal load characteristic classification according to an embodiment of the present invention (hereinafter referred to as the "sterilizing dryer") includes a memory for storing instructions and a processor for executing instructions loaded from the memory. Here, the instructions cause the processor to perform a sterilizing dryer control method according to an embodiment of the present invention. According to the embodiment, the function performed at each step is converted into a respective configuration and can be implemented as a device including these configurations. Since each configuration overlaps with the description of each step mentioned above, a detailed description is omitted.

[0076] The contents described in the description of the invention are merely illustrative, and the invention may be modified and implemented in various ways by those skilled in the art. Accordingly, the scope of protection of the invention is not limited to the description and expression of the described embodiments. Explanation of the symbols

[0077] 100: Sterilizing dryer 110: Sensor section 120: Heating part 130: Ventilation unit 140: Control unit 150: Sterilization section

Claims

Claim 1 A method performed in a processor controlling a sterilization dryer comprises: a step of generating temperature data within a chamber; a step of generating humidity data within the chamber; a step of generating a load vector including a rate of change in temperature and a rate of change in humidity per unit time using the temperature data and the humidity data; a step of inputting the load vector into a load classification model, which is an artificial intelligence model, to classify a load class for the thermal load characteristics of an object to be dried within the chamber or the operating state of the sterilization dryer; and a step of controlling a heater according to a control profile corresponding to the load class, wherein the step of generating the load vector and the step of controlling the heater are repeated at a predetermined period, and the step of generating the load vector comprises calculating the rate of change in temperature within the chamber during a first period until the temperature within the chamber reaches a predetermined target temperature and the rate of change in humidity within the chamber during a second period from when the temperature within the chamber reaches the target temperature until sterilization drying is completed, and combining the rates of change at different points in time into a single vector, and the step of generating the load vector comprises a step of calculating the rate of change in temperature during the first period;The method comprises the step of calculating the humidity change rate at each predetermined period in the second section and updating the load vector by combining it with the temperature change rate, wherein the step of classifying the load class comprises inputting the updated load vector into the load classification model at each predetermined period, and the step of controlling the heater comprises, if the load class is classified into a first class distributed in an area having a relatively high temperature change rate and a relatively low humidity change rate in the feature space of the load vector, setting the output ratio of the heater lower than a preset reference output ratio and setting the operating time of the heater shorter than a preset time; if the load class is classified into a third class distributed in an area having a relatively low temperature change rate and a humidity change rate in the feature space, setting the operating time of the heater shorter than a preset time; and if the load class is classified into a fourth class distributed in an area having a relatively low temperature change rate and a relatively high humidity change rate in the feature space, setting the output ratio of the heater higher than a preset reference output ratio or to a maximum value. Claim 2 delete Claim 3 delete Claim 4 A method for controlling an artificial intelligence sterilization dryer based on thermal load characteristic classification according to claim 1, wherein the step of generating the load vector is characterized by calculating the rate of change of temperature per unit time based on the time required to reach the target temperature by dividing the rate of change of temperature when the inside of the chamber is empty. Claim 5 A method for controlling an artificial intelligence sterilization dryer based on thermal load characteristic classification according to claim 1, wherein the load classification model is a machine learning model trained by supervised learning using learning data that includes input values ​​including temperature change rates and humidity change rates collected in the first section and the second section, and output values ​​including load classes classified based on the actual amount of material to be dried and drying state under the same conditions. Claim 6 A method for controlling an artificial intelligence sterilization dryer based on thermal load characteristic classification according to claim 1, wherein the load classification model is a model learned through unsupervised learning based on accumulated distribution data of the load vector, and is characterized by defining and classifying each cluster distinguished based on similarity between the load vectors as a load class in the feature space of the load vector formed by the temperature change rate and the humidity change rate. Claim 7 A method for controlling an artificial intelligence sterilizing dryer based on thermal load characteristic classification according to claim 1, wherein the load classification model is a model learned through reinforcement learning by defining the load vector as a state and the selection of the load class as a behavior, and parameters are updated in a direction that maximizes a reward calculated based on the drying success rate according to whether the target humidity is reached and the energy efficiency according to the amount of power input, as a result of performing a control profile according to the selected load class. Claim 8 delete Claim 9 A method for controlling an artificial intelligence sterilizing dryer based on thermal load characteristic classification according to claim 1, wherein the step of classifying the load class is to classify it into a fifth class regardless of the humidity change rate when the temperature change rate is less than a predetermined threshold, and the step of controlling the heater is to perform intermittent operation in which, when classified into the fifth class, the operating time of the heater is set to a restart time shorter than a preset time, and the heater is operated again for the restart time after a preset waiting time following the shutdown of the heater. Claim 10 An artificial intelligence sterilizing dryer based on thermal load characteristic classification, comprising: a sensor unit for detecting temperature and humidity within a chamber; a heating unit including a heater for heating the chamber; and a control unit including a memory for storing instructions and a processor for executing instructions loaded from the memory, wherein the instructions cause the processor to perform the method of any one of claims 1, 4 to 7, and 9.

Images