Method and system for odor detection and monitoring.

Abstract

The present invention relates to an odor detection and monitoring method and system that uses machine learning techniques to predict the type and intensity of an odor from an odor measured by at least one gas sensor. The present invention involves at least one panelist evaluating and labeling the odor intensity of several odor samples, and using the measured gas sensor data of the same odors evaluated by the panelist to train one or more odor analysis models for machine learning. The trained models are used to predict the type and intensity of the odor.

Description

[Technical Field]

[0001] The present invention relates to a method and system for odor detection and monitoring using a gas sensor. This method and system is particularly suitable for measuring odors in target areas such as chemical product manufacturing lines, waste recycling facilities, and water treatment facilities. [Background technology]

[0002] As the world's population grows, industrial and residential areas are becoming closer together, increasing the importance of odor detection and monitoring. In most countries, residential areas expand toward industrial areas. Therefore, it becomes difficult for industries to control odor emissions without affecting surrounding areas. The type and intensity of odors are important parameters that should be monitored. Knowing only one of them is not sufficient, especially when warnings must be communicated to residents in residential areas. Therefore, a system is needed that can accurately determine the type and intensity of odors emitted from industrial areas. Furthermore, this system must provide the type and intensity of detected odors, at least in accordance with conventional olfactory techniques that must be evaluated by certified panelists. Certified panelists are experts who perform odor sensory evaluations. Those who wish to become certified panelists must undergo training and obtain qualifications. In order to maintain appropriate evaluation results, the evaluation steps for each odor sensor must be carefully performed. Acceptable standards that must be followed vary from country to country. Standard methods for smelling odor samples are recommended based on those of the American Society for Testing and Materials (ASTM) and the Japanese Industrial Standard (JIS). Portable olfactory techniques according to the EN13725:2022 standard are another industrially acceptable standard. Therefore, the objective of this invention is to provide a system capable of detecting odor type and intensity at the same level as panelists.

[0003] International Publication No. 2022005407 by the same inventors discloses an odor detection and monitoring system. This system comprises an odor sensor that collects odor samples from the air and analyzes the odor composition from the samples, a weather station, and a processing system that determines the type and diffusion pattern of the odor. The system displays odor diffusion information on a geographical image, allowing the user to visualize the odor diffusion pattern. In this disclosure, only one odor sensor is used to collect and analyze the odor composition. If the sensor does not react to a particular odor composition, accurate analysis is not provided. Furthermore, this system is intended to analyze odor concentration for pattern prediction, not to analyze odor intensity. Therefore, standardized olfactory techniques cannot be applied to the results.

[0004] U.S. Patent No. 9028751 discloses a system and method for controlling odors emitted from an odor source or factory. This invention uses an odor sensor to detect odor emissions. Sensor data is processed in combination with weather information to analyze odor diffusion. If the odor intensity is found to be above a specified level, the system reduces odor emissions until the odor intensity falls below the specified level. This system is intended for use in detecting known odors. It cannot respond to odors not supported by the sensor.

[0005] U.S. Patent No. 1,111,2383 and its corresponding application describe odor detection systems and methods that utilize multiple sensors or array sensors to detect odors. At least two sensors (or sensor elements of an array) interact with the odor composition of a measured odor sample. The odor concentration can be measured by varying the detection sensitivity of each sensor. An information processing unit is then used to perform pattern recognition based on sensor characteristics such as sensor type and arrangement. The present invention uses the array sensor response of different types of sensors to identify the type of detected odor, but it cannot measure or predict the intensity of the detected odor as needed. [Prior art documents] [Patent Documents]

[0006] [Patent Document 1] International Publication No. 2022005407 [Patent Document 2] U.S. Patent No. 9028751 [Patent Document 3] U.S. Patent No. 11112383 [Overview of the project] [Problems that the invention aims to solve]

[0007] Therefore, there is still a need to provide a system and / or method that can accurately detect or predict both the type and intensity of odors. In particular, the prediction results should conform to at least conventional olfactory techniques without requiring panelist testing. This would allow the system to be used in a variety of factories and industries. [Means for solving the problem]

[0008] The present invention relates to a method for odor detection and monitoring. The method includes (a) creating an odor database used to store a set of sample odor data for training an odor analysis model, wherein the data in the set includes at least the type of odor, the intensity of the odor, temperature, humidity, and measured gas sensor data from at least one gas sensor from odor data measurements; (b) creating a plurality of odor analysis models for processing the measured gas sensor data together with temperature and humidity data in order to predict the type and intensity of the odor measured by at least one gas sensor; and (c) analyzing the measured gas sensor data in order to predict the type and intensity of the odor from the odor measured by at least one gas sensor, wherein at least one odor analysis model is used to process the measured gas sensor data together with temperature and humidity data by using machine learning techniques. The present method is characterized in that (a) the step of creating an odor database includes at least the steps of (a-1) having at least one panelist evaluate and label the type of odor of an odor sample, (a-2) having the panelist evaluate and label the intensity of the odor of the odor sample, (a-3) storing measured gas sensor data from at least one sensor used to measure the same odor as the odor evaluated by the panelist, (a-4) storing ambient temperature and humidity data when the sensor measures the odor, (a-5) providing a set of sample odor data, each set including at least the type of odor labeled by the panelist, the intensity of the odor labeled by the panelist, measured gas sensor data, temperature data, and humidity data, and (a-6) storing the set of sample odor data in the odor database.

[0009] In another embodiment, the step of creating a plurality of odor analysis models may include at least (b-1) preprocessing the set of sample odor data acquired in step (a), (b-2) transforming the preprocessed dataset, (b-3) training at least one odor analysis model using the transformed dataset, (b-4) evaluating the performance of the trained odor analysis model, and (b-5) storing the trained odor analysis model and its parameters.

[0010] In another embodiment, the step of (c) analyzing the measured gas sensor data may include at least (c-1) storing the measured gas sensor data from at least one sensor, (c-2) storing ambient temperature and humidity data when the sensor measures the odor, (c-3) processing the measured gas sensor data together with the temperature and humidity data by using at least one trained odor analysis model, and (c-4) predicting the type and intensity of the odor measured by the sensor.

[0011] In a preferred embodiment, (a-2) the step of a panelist evaluating and labeling the odor intensity of an odor sample can be selected from (1) collecting the odor sample in an odor bag and analyzing the odor by smelling the odor sample according to the methods of the American Society for Testing and Materials (ASTM) or the Japanese Industrial Standards (JIS), or (2) smelling the odor in the environment using a portable olfactory technique in accordance with the EN13725:2022 standard.

[0012] In another preferred embodiment, the odor analysis model may include an odor type prediction model and an odor intensity prediction model. In some embodiments, the step of (b) creating multiple odor analysis models further includes the step of monitoring the deviations between the odor analysis models, thereby allowing the models to be evaluated and their parameters to be adjusted.

[0013] In another embodiment, a method for odor detection and monitoring may further include the step of displaying odor monitoring results to show at least measured gas sensor data from at least one sensor, predicted odor type, predicted odor intensity, temperature, and humidity. Additional data to be displayed may be further selected from odor type prediction models, odor intensity prediction models, radar charts representing odor composition, gas sensor status, gas sensor location, local weather data, odor diffusion patterns, and a history of odor monitoring data.

[0014] The present invention also relates to a system for odor detection and monitoring. The system comprises an odor database, at least one gas sensor, a temperature sensor, a humidity sensor, a processing unit, and a user interface unit, wherein the odor database stores data used for training an odor analysis model and analyzing odors; the gas sensor measures odor samples diffused in the air within an odor monitoring area and transmits the measured gas sensor data to the processing unit via communication means; the temperature sensor collects ambient outside air temperature data when the sensor measures odors; the humidity sensor collects ambient outside air humidity data when the sensor measures odors; the processing unit performs data processing to create an odor analysis model and analyze odors; and the user interface unit displays data to the user or receives input from the user. This system is characterized in that the odor database is divided into at least two datasets, comprising a training dataset for an odor analysis model and a dataset for the odor analysis model, wherein the training dataset for the odor analysis model comprises the type and intensity of odors labeled by at least one panelist, measured gas sensor data from at least one sensor used to measure the same odor sample labeled by the panelist, and temperature and humidity data during measurement, and the odor analysis model dataset comprises at least one odor analysis model and its parameters. The processing unit uses at least one odor analysis model to process the measured gas sensor data together with the temperature and humidity data by using machine learning techniques.

[0015] In some embodiments, the system's processing unit may further include a filtering unit for filtering the measured gas sensor data and / or a normalization unit for normalizing the measured gas sensor data by using baseline data obtained from clean air measurements.

[0016] In a preferred embodiment, the user interface unit includes at least a portion for displaying the measured gas sensor data of at least one sensor, a portion for displaying the predicted odor type data, a portion for displaying the predicted odor intensity data, a portion for displaying the temperature sensor data, and a portion for displaying the humidity sensor data.

[0017] In another preferred embodiment, the user interface unit may further include a data display portion for displaying data selectable from an odor type prediction model, an odor intensity prediction model, a radar chart representing the odor composition, the state of the gas sensor, the position of the gas sensor, the meteorological data of the region, the predicted odor diffusion pattern, and the odor monitoring data history.

[0018] In some embodiments, the user interface unit further includes a portion for displaying the deviation monitoring of the odor analysis model, whereby the model can be evaluated or its parameters can be adjusted.

[0019] The object of the present invention is to provide an odor detection and monitoring method that uses machine learning techniques to predict the type and intensity of an odor from an odor measured by at least one gas sensor. The method involves at least one panelist evaluating and labeling the odor intensity of several odor samples, and training one or more odor analysis models for machine learning using the measured gas sensor data of the same odors evaluated by the panelist. The trained models are then used to analyze the gas sensor data measured from at least one sensor to predict the type and intensity of the odor. Preferred steps of the method include the panelist evaluating and labeling the odor intensity of the odor samples based on (1) collecting the odor samples in an odor bag and analyzing the odor by smelling the odor samples according to the American Society for Testing and Materials (ASTM) or Japanese Industrial Standards (JIS) methods, or (2) smelling the odor in the environment using a portable olfactory technique according to the EN13725:2022 standard.

[0020] Another object of the present invention is to provide a system for odor detection and monitoring based on the above method. The system includes an odor database, at least one gas sensor, a temperature sensor, a humidity sensor, a processing unit, and a user interface unit. The odor database stores data used for training an odor analysis model and analyzing odors. The gas sensor measures an odor sample diffused in the air within an odor monitoring area, and transmits the measured gas sensor data to the processing unit via communication means. On the other hand, the temperature sensor and the humidity sensor respectively collect ambient outside air temperature data and humidity data. The processing unit performs data processing to create an odor analysis model and analyze odors. The system is characterized in that the odor database is divided into at least two data sets, comprising a training data set for the odor analysis model and a data set for the odor analysis model. The training data set for the odor analysis model includes the type and intensity of odors labeled by at least one panelist, the measured gas sensor data of at least one sensor used to measure the same odor sample labeled by the panelist, and the temperature data and humidity data during measurement. The odor analysis model data set includes at least one odor analysis model and its parameters. The processing unit uses at least one odor analysis model to process the measured gas sensor data together with the temperature data and humidity data by using machine learning techniques to predict the type and intensity of odors.

Brief Description of the Drawings

[0021] [Figure 1] It is a diagram showing a method for odor detection and monitoring of the present invention. [Figure 2] It is a diagram showing steps for creating an odor database. [Figure 3] It is a diagram showing steps for creating a plurality of odor analysis models. [Figure 4] It is a diagram showing steps for analyzing measured gas sensor data. [Figure 5] This is a radar chart of odor characteristics measured by five different types of gas sensors. [Figure 6] This is a schematic block diagram of the odor detection and monitoring system of the present invention. [Figure 7] This figure shows an example of a user interface unit. [Figure 8A] This is a radar chart that shows the characteristics of the odor. [Figure 8B] This is another radar chart that represents the characteristics of odors. [Figure 8C] This is another radar chart that represents the characteristics of odors. [Figure 9] This figure shows an example of the predictive performance of a machine learning model. [Modes for carrying out the invention]

[0022] Figures 1 and 2 illustrate the method for odor detection and monitoring according to the present invention. The method includes steps 10 to create an odor database, 20 to create multiple odor analysis models, and 30 to analyze measured gas sensor data to predict the type and intensity of unknown odors. To analyze the measured gas sensor data, it is necessary to create an appropriate odor analysis model. Machine learning algorithms can be trained and tested to evaluate their performance. High-accuracy algorithms may be selected as odor analysis models. Note that multiple models can be used in the prediction process. Results predicted based on multiple models provide a higher level of confidentiality compared to results from a single model.

[0023] To create a model, the odor database is used to store a set of sample odor data for training the odor analysis model. The data in the set includes at least the type of odor, the intensity of the odor, temperature, humidity, and measured gas sensor data from at least one gas sensor from the odor data measurement.

[0024] Since the sample odor dataset is used to train the odor analysis model, its accuracy is extremely important to this invention. If the dataset is not accurate, the trained model will provide inaccurate predictions. On the other hand, if an accurate dataset is used for training, the model will provide accurate predictions.

[0025] In this invention, preferably, a certified panelist is used to create a set of sample odor data. A certified panelist is an expert who performs odor sensory evaluation. The requirements for becoming a certified panelist may vary from country to country. In this invention, a certified panelist is required to evaluate odor samples. The evaluation must include at least the determination of the type and intensity of the odor sample. Simultaneously, odor samples are also collected or measured by a gas sensor to create measurement data. This method allows the measured data to be compared with the panelist's evaluation. This comparison is further used to train an odor analysis model.

[0026] To create an odor database, the following steps are typically performed: (i) step 11 in which at least one panelist evaluates and labels the type of odor of an odor sample; (ii) step 12 in which the panelist evaluates and labels the intensity of the odor of the same odor sample; (iii) step 13 in which the measured gas sensor data of at least one sensor used to measure the same odor as the odor evaluated by the panelist is stored; (iv) step 14 in which ambient temperature and humidity data is stored when the sensor measures the odor; (v) step 15 in which a set of sample odor data is provided; and (vi) step 16 in which the set of sample odor data is stored in the odor database. The set of sample odor data must include at least the type and intensity of the odor labeled by the panelist, the measured gas sensor data, and the temperature and humidity data.

[0027] To create an analytical model used to predict the type and intensity of odors measured by at least one gas sensor, the sample odor dataset is analyzed using machine learning techniques. At least one odor analysis model is used to provide predictions for the measured gas sensor data, temperature data, and humidity data. Multiple models can be used to improve predictive performance.

[0028] Many possible odor analysis models can be applied to the present invention. To obtain an accurate model, it is necessary to properly pre-process the set of sample odor data for model training so that the resulting model parameters match the provided data. Therefore, when creating an odor analysis model as shown in Figure 3, it is preferable to perform the following steps: (i) step 21 of pre-processing the set of sample odor data; (ii) step 22 of transforming the pre-processed dataset; (iii) step 23 of training the odor analysis model using the transformed dataset; (iv) step 24 of evaluating the performance of the trained odor analysis model; and (v) step 25 of storing the trained odor analysis model and its parameters. Various models can be pre-processed in the same way to create a model that is best suited for further analysis. These steps can be performed on multiple models simultaneously or sequentially in a loop.

[0029] Examples of preprocessing steps include data cleaning, correction of outliers, removal of outliers, filtering of sensor data, and scaling. Transformation steps may include data aggregation, normalization, and generalization.

[0030] After the model is trained to meet its designed predictive performance, it can be used to analyze gas sensor data measured from actual measurements to predict the type and intensity of unknown odors in the monitoring area. The analysis step uses the trained model to predict the type and intensity of odors without evaluation information from panelists. This is the situation in which the present invention is utilized in a real environment. Since the model is trained on information evaluated by panelists, it is expected to provide predictions of odor types and intensity as accurately as the panelists themselves.

[0031] The analysis step preferably includes, as shown in Figure 4, (i) step 31 of storing measured gas sensor data from at least one sensor; (ii) step 32 of storing ambient temperature and humidity data when the sensor measures odor; (iii) step 33 of processing the measured gas sensor data together with the temperature and humidity data by using at least one trained odor analysis model; and (iv) step 34 of predicting the type and intensity of the odor measured by the sensor.

[0032] The predictive performance of this invention depends on the accuracy of the sample odor dataset stored in the database. Since the dataset is used to train the odor analysis model, the model's predictive accuracy follows the accuracy of the prepared dataset. If the dataset is correctly labeled, the model's accuracy will be high. On the other hand, if there are errors in the labeled dataset, the model's accuracy will be low. It is well understood in the machine learning community that the accuracy of the training dataset must be as high as possible to obtain an accurate model. This invention uses panelists to evaluate and label the odor types and intensities of the training dataset, thus ensuring that the accuracy of the dataset is as high as that of standard olfactory methods, resulting in an accurate and standardized odor analysis model.

[0033] The gas sensor is preferably a metal oxide gas sensor that is sensitive to the main characteristics of the target odor. Preferably, the sensor is a hydrocarbon (C) X H Y ), hydrogen gas (H2), carbon monoxide (CO), alcohol (C) X H Y The odor must be one of the following: OH, nitrogen dioxide (NO2), nitric oxide (NO), ozone (O3), ammonia (NH3), sulfur dioxide (SO2), refrigerant R134a, other volatile organic compounds, or a combination thereof. Multiple sensors can be used to measure odor samples and distinguish between various odors. Typically, the amplitude of the sensor signal or data represents the intensity of the detected odor.

[0034] Figure 5 is a radar chart of odor characteristics measured by five different gas sensors. Each corner of the chart indicates the amplitude level of the detected gas. Preferably, the measured gas sensor data is acquired from an array sensor with eight or more sensors to create a distinguishable pattern on the radar chart. It is clear that odor characteristics can be determined by using different types of sensors. The odor characteristics, odor type, and intensity, labeled by the panelists, are grouped together as a training dataset used in the training process of the odor analysis model.

[0035] In a preferred embodiment, the panelist evaluates and labels the types and intensities of odors in the training dataset by either (1) collecting odor samples in an odor bag and analyzing the odors by smelling the samples according to the methods of the American Society for Testing and Materials (ASTM) or the Japanese Industrial Standards (JIS), or (2) smelling odors in the environment using a portable olfactory technique in accordance with the EN13725:2022 standard. These methods or standards are accepted worldwide in odor assessment. Therefore, the present invention aims to train a machine learning model to achieve the same level of accuracy as the standards. This is particularly useful to reduce erroneous predictions based on human error or to enhance the usefulness of the present invention for odor monitoring when a panelist is absent. Once created, the training dataset can be reused to train the odor analysis model any number of times. The training dataset may be updated from time to time to improve the accuracy of predictions or to increase the number of detectable odor types.

[0036] In some embodiments, odor type and intensity can be grouped as two dependent variables, so a single odor analysis model may be selected and trained to predict both odor type and intensity. In other embodiments, the odor analysis model preferably comprises an odor type prediction model and an odor intensity prediction model. In such cases, one prediction model is used to predict one variable. By using the two models separately, the prediction accuracy for each variable, i.e., odor type or odor intensity, is improved. The prediction models may be the same or may be completely different.

[0037] Many machine learning models can be used to predict odor types. For example, they include decision trees, random forests, naive Bayes, support vector machines, K-nearest neighbors (KNN), gradient boosting machines, light gradient boosting, extreme gradient boosting, extreme randomized tree models, distributed random forests, generalized linear models, stack ensembles, deep learning, multilayer perceptrons, recurrent neural networks, long short-term memory networks, generative adversarial networks, and autoencoder deep learning algorithms. Preferably, the odor type prediction model is an extreme randomized tree model.

[0038] Similarly, many other machine learning models can be used to predict odor intensity. These include linear regression, logistic regression, polynomial regression, Lasso regression, Bayesian linear regression, principal component regression, partial least squares regression, and elastic net regression. Preferably, the odor intensity prediction model is polynomial regression.

[0039] In some embodiments, it is preferable to include a monitoring step for monitoring deviations in the odor analysis model in order to facilitate the creation of the odor analysis model. By observing deviations in the model, such as accuracy, precision, and recall, it is possible to identify when the model's predictive performance has deteriorated and to determine whether the model needs to be retrained to improve its accuracy.

[0040] In some embodiments, the present invention may include the step of setting a warning when the odor intensity level exceeds a threshold. To provide continuous odor monitoring, the present invention must be able to warn the user when the intensity of a particular odor exceeds a predefined threshold, for example, 300-1,000 OU / m 3 The odor intensity is the threshold for discharge from the chimney, which is 15-30 OU / m³. 3 The intensity of the odor is limited by the surrounding conditions.

[0041] When analyzing measured gas sensor data, the sensor data may be filtered to reduce noise within the data. Depending on the type and accuracy of the sensor, various filtering techniques may be applied, such as low-pass filtering, band-pass filtering, or high-pass filtering. Preferably, filtering is performed by using an exponential filter that smooths noisy measurements while minimizing memory usage. Appropriate filter parameters can be stored in the odor database and applied to the sensor data as needed. Parameters can also be adjusted by the user to achieve the best predictive performance. Similarly, it is understandable to those skilled in the art that other processing or signal processing techniques, such as upsampling / downsampling or transformation, may be applied to the sensor data before and / or after the filtering process.

[0042] In a preferred embodiment, the measured gas sensor data can be normalized by using baseline data obtained from clean air measurements. Clean air is odorless, pure, and free of odor particles. Using clean air data as a reference is equivalent to calibrating the sensor to have a zero offset relative to clean air. Therefore, the accuracy of the predictive performance of the present invention is improved.

[0043] In some embodiments, the method for odor detection and monitoring according to the present invention may include step 40 of displaying odor monitoring results to show at least measured gas sensor data from at least one sensor, predicted odor type, predicted odor intensity, and temperature and humidity of the monitored area.

[0044] In another embodiment, additional information may be included when displaying odor monitoring results. This information should be selected to provide the user with an overall scenario of odor monitoring. Preferably, the information may be selected from an odor type prediction model, an odor intensity prediction model, a radar chart representing the odor composition, the status of the gas sensor, the location of the gas sensor, local weather data, the odor diffusion pattern, and a history of odor monitoring data.

[0045] Next, the odor detection and monitoring system 100 will be described in conjunction with the above method. Figure 6 is a schematic block diagram of the odor detection and monitoring system 100 according to the present invention. The system comprises an odor database 110, at least one gas sensor 120, a temperature sensor 130, a humidity sensor 140, a processing unit 150, and a user interface unit 160. The odor database 110 stores data used for training an odor analysis model and for analyzing odors measured in the odor monitoring area. The odor database 110 is divided into at least two datasets, comprising an odor analysis model training dataset and an odor analysis model dataset. The odor analysis model training dataset comprises the type and intensity of odors labeled by at least one panelist, measured gas sensor data from at least one sensor used to measure the same odor sample labeled by the panelist, temperature data and humidity data during measurement. The odor analysis model dataset comprises at least one odor analysis model and its parameters. The gas sensor 120 measures an odor sample diffused in the air within the odor monitoring area and transmits the measured gas sensor data to the processing unit 150 via communication means. The communication means or medium may be wired or wireless, and may be either a local or network connection. The temperature sensor 130 collects ambient air temperature data when the gas sensor 120 measures the odor sample. The humidity sensor 140 collects ambient air humidity data when the gas sensor 120 measures the odor sample. The processing unit 150 performs data processing to analyze the odor by creating an odor analysis model and processing the measured gas sensor data together with the temperature and humidity data using machine learning techniques in at least one odor analysis model. The user interface unit 160 displays data to the user or receives input from the user. The user interface unit 160 may display measured and processed data, including a graphical display of sensor data.

[0046] In a preferred embodiment, the processing unit 150 performs data processing to create an odor analysis model, by having at least the steps of (i) pre-processing a training dataset for an odor analysis model, (ii) transforming the pre-processed dataset, (iii) training at least one odor analysis model, (iv) evaluating the performance of the trained odor analysis model, and (v) storing the trained odor analysis model and its parameters.

[0047] To analyze the measured odor, the processing unit performs data processing in accordance with at least the following steps: (i) storing measured gas sensor data from at least one sensor 120; (ii) storing temperature data and humidity data; (iii) processing the measured gas sensor data from at least one sensor 120 together with the temperature data and humidity data, as well as at least one odor analysis model from an odor analysis model dataset; and (iv) predicting the type and intensity of the odor in the measured data.

[0048] The types and intensities of odors stored in the odor database are preferably labeled by the panelists to ensure the accuracy of odor predictions by the odor analysis model. Panelists may choose one of the following methods for labeling odors: (1) collecting odor samples in an odor bag and analyzing the odor by smelling the samples according to the methods of the American Society for Testing and Materials (ASTM) or the Japanese Industrial Standards (JIS); or (2) smelling the odors in the environment using a portable olfactory technique in accordance with the EN13725:2022 standard. These two odor determination methods are well known and widely accepted as industrial standards.

[0049] In some embodiments, the odor analysis model can be divided into two different models: an odor type prediction model and an odor intensity prediction model. When high prediction accuracy is required, it is recommended to use two models dedicated to specific prediction purposes. However, in some embodiments, only one prediction model may be applied to predict both odor type and intensity.

[0050] Many machine learning models can be applied to predict the type of odor. Odor type prediction models can be selected from decision trees, random forests, naive Bayes, support vector machines, K-nearest neighbors (KNN), gradient boosting machines, light gradient boosting, extreme gradient boosting, extreme randomized tree models, distributed random forests, generalized linear models, stack ensembles, deep learning, multilayer perceptrons, recurrent neural networks, long short-term memory networks, generative adversarial networks, and autoencoder deep learning algorithms. Preferably, the odor type prediction model is an extreme randomized tree model.

[0051] Similarly, various machine learning models can be used to predict odor intensity. Odor intensity prediction models can be selected from linear regression, logistic regression, polynomial regression, Lasso regression, Bayesian linear regression, principal component regression, partial least squares regression, and elastic net regression. Preferably, the odor intensity prediction model is polynomial regression.

[0052] In some embodiments, the processing unit 150 may further include a filtering unit for filtering the measured gas sensor data. The filtering unit is designed to reduce the noise of the sensor data, improving the prediction performance of the system. Depending on the characteristics of the measured sensor data, various filtering units may be selected, such as low-pass filtering, band-pass filtering, or high-pass filtering. Preferably, the filtering unit is an exponential filter because it requires less memory to store data. It will be understandable to those skilled in the art that other filtering or signal processing units may be applied to the sensor data before and / or after the filtering unit, such as upsampling / downsampling or conversion.

[0053] In some embodiments, the processing unit 150 may further include a normalization unit for normalizing the measured gas sensor data by using baseline data obtained from clean air measurements. By using the clean air data as reference data, the accuracy of the prediction performance of the system is improved.

[0054] The gas sensor 120 is preferably a metal oxide gas sensor from which the measured gas sensor data is obtained. They can be sensors that can measure hydrocarbons (C X H Y ), hydrogen gas (H2), carbon monoxide (CO), alcohol (C X H Y OH), nitrogen dioxide (NO2), nitric oxide (NO), ozone (O3), ammonia (NH3), sulfur dioxide (SO2), refrigerant R134a, other volatile organic compounds, or combinations thereof. For measuring odor samples, multiple sensors are preferred. Preferably, the gas sensor is an array sensor having eight or more sensors capable of measuring the above-mentioned gases.

[0055] In a preferred embodiment, the user interface unit 160 includes at least a portion for displaying measured gas sensor data from at least one sensor, a portion for displaying predicted odor type data, a portion for displaying predicted odor intensity data, a portion for displaying temperature sensor data, and a portion for displaying humidity sensor data. Figure 7 shows an example of the user interface unit 160 according to the present invention.

[0056] Additional information may be selected to be displayed by the user interface unit 160 to represent information useful to the user. The user interface unit 160 may include a data display section for displaying an odor type prediction model, an odor intensity prediction model, a radar chart representing the odor composition, the status of the gas sensor, the location of the gas sensor, local weather data, the predicted odor diffusion pattern, and / or the odor monitoring data history.

[0057] In some embodiments, the user interface unit 160 may further include a section for displaying deviation monitoring of the odor analysis model, thereby allowing the user to monitor and evaluate the analysis model. The user interface unit (160) may also further include a section for setting and displaying a warning when the odor intensity level exceeds a threshold. [Examples]

[0058] The present invention was tested by the inventors to verify its accuracy. Three array sensors, each having 10 sensor elements, were installed in the paper mill to monitor odors in various areas within the mill. These were (a) the papermaking machinery, (b) the biofilter, and (c) the wastewater treatment area. Each area contained different odors, each with its own unique characteristics, and was distinguishable as shown in the radar charts in Figures 8A-8C. The measured odor intensity was predicted. The odor intensity in the wastewater treatment area was 1,000-2,500 OU / m³. 3The odor intensity in the biofilter area was 2,000-7,000 OU / m³. 3 The odor intensity in the papermaking machinery was 500-2,000 OU / m³. 3 The digital data from the sensors was divided into two groups. The first group was data from the sensors, and the second group was environmental data including temperature, humidity, sensor location, and weather information. Several machine learning models were used to verify the present invention. The processing unit performed data analysis, data cleaning, feature selection, and feature engineering for various machine learning models. The processing unit performed outlier detection to remove anomalies from the data groups. The sensor data was filtered to reduce noise and normalized using baseline data, which were clean air measurements at its location. The sensor data was then grouped by using a hierarchical clustering technique to eliminate redundancy, and important feature groups of the data were selected. Machine learning models were trained, and their predictive accuracy was evaluated. The optimal model was found to be a polynomial regression model, and the results are shown in Figure 9.

[0059] The inventors have conducted similar verifications to validate the performance of the present invention in various factory scenarios, such as seafood processing plants and fresh food packaging plants. However, to conserve space in this disclosure, those results are not described herein.

Claims

1. A method for odor detection and monitoring, (a) A step (10) of creating an odor database used to store a set of sample odor data for training an odor analysis model, wherein the data in the set includes at least the type of odor, the intensity of the odor, temperature, humidity, and measured gas sensor data from at least one gas sensor, (b) Step (20) to create multiple odor analysis models for processing the measured gas sensor data together with temperature data and humidity data in order to predict the type and intensity of odor measured by at least one gas sensor, (c) A step (30) of analyzing measured gas sensor data in order to predict the type and intensity of the odor from the odor measured by at least one gas sensor, wherein at least one odor analysis model is used to process the measured gas sensor data together with temperature data and humidity data by using machine learning techniques, In a method for odor detection and monitoring, including, (a) Step (10) of creating the odor database, (a-1) A step (11) in which at least one panelist evaluates and labels the type of odor in the odor sample, (a-2) The panelists evaluate the odor intensity of the odor sample and label it (12), (a-3) Step (13) of storing measured gas sensor data from at least one sensor used to measure the same odor as the odor evaluated by the panelists, (a-4) Step (14) of storing ambient outside air temperature data and humidity data when the sensor measures odor, (a-5) Step (15) of providing a set of sample odor data, wherein each set includes at least the type of odor labeled by the panelist, the intensity of the odor labeled by the panelist, the measured gas sensor data, the temperature data, and the humidity data. (a-6) Step (16) of storing the set of sample odor data in the odor database, A method for odor detection and monitoring, characterized by comprising at least [a certain element].

2. (b) The step (20) of creating the plurality of odor analysis models, (b-1) Step (21) of pre-processing the set of sample odor data obtained in step (a), (b-2) Step (22) of transforming the pre-processed dataset, (b-3) Step (23) of training at least one odor analysis model using the converted dataset, (b-4) Step (24) to evaluate the performance of the trained odor analysis model, (b-5) Step (25) of storing the trained odor analysis model and its parameters, A method for odor detection and monitoring according to claim 1, comprising at least the following:

3. (c) Step (30) of analyzing the measured gas sensor data, (c-1) Step (31) to store measured gas sensor data from at least one sensor, (c-2) Step (32) of storing ambient temperature data and humidity data when the sensor measures odor, (c-3) The step (33) of processing the measured gas sensor data together with the temperature data and humidity data by using at least one trained odor analysis model, (c-4) A step (34) to predict the type of odor and the intensity of the odor measured by the sensor, A method for odor detection and monitoring according to claim 1, comprising at least the following:

4. (a-2) The method for odor detection and monitoring according to claim 1, wherein the step (12) of the panelists evaluating and labeling the odor intensity of the odor sample is selected from (1) collecting the odor sample in an odor bag and then analyzing the odor by smelling the odor sample according to the methods of the American Society for Testing and Materials (ASTM) or the Japanese Industrial Standards (JIS), or (2) smelling the odor in the environment using a portable olfactory technique in accordance with the EN13725:2022 standard.

5. The method for odor detection and monitoring according to claim 1, wherein the odor analysis model comprises an odor type prediction model and an odor intensity prediction model.

6. The method for odor detection and monitoring according to claim 5, wherein the odor type prediction model can be selected from decision trees, random forests, naive Bayes, support vector machines, K-nearest neighbors (KNN), gradient boosting machines, light gradient boosting, extreme gradient boosting, extreme randomized tree models, distributed random forests, generalized linear models, stacked ensembles, deep learning, multilayer perceptrons, recurrent neural networks, long short-term memory networks, generative adversarial networks, and autoencoder deep learning algorithms.

7. The method for odor detection and monitoring according to claim 5, wherein the odor type prediction model is an extreme randomization tree model.

8. The method for odor detection and monitoring according to claim 5, wherein the odor intensity prediction model can be selected from linear regression, logistic regression, polynomial regression, Lasso regression, Bayesian linear regression, principal component regression, partial least squares regression, and elastic net regression.

9. The method for odor detection and monitoring according to claim 5, wherein the odor intensity prediction model is polynomial regression.

10. (b) The method for odor detection and monitoring according to claim 1, wherein the step (20) of creating the plurality of odor analysis models further includes the step of monitoring the deviation of the odor analysis models.

11. A method for odor detection and monitoring according to claim 1, further comprising the step of setting a warning when the intensity level of the odor exceeds a threshold.

12. (c) The method for odor detection and monitoring according to claim 1, wherein the step (30) of analyzing the measured gas sensor data includes at least the step of filtering the measured gas sensor data.

13. The method for odor detection and monitoring according to claim 12, wherein the filtering step is performed by using an exponential filter.

14. (c) The method for odor detection and monitoring according to claim 1, wherein the step (30) of analyzing the measured gas sensor data includes at least the step of normalizing the measured gas sensor data by using baseline data obtained from a clean air measurement.

15. The method for odor detection and monitoring according to claim 1, wherein the measured gas sensor data is obtained from a metal oxide gas sensor.

16. The measured gas sensor data is hydrocarbon (C Ｘ H Ｙ ), hydrogen gas (H ２ ), carbon monoxide (CO), alcohol (C Ｘ H Ｙ OH), nitrogen dioxide (NO ２ ), nitric oxide (NO), ozone (O ３ ), ammonia (NH ３ ), sulfur dioxide (SO ２ ), refrigerant R134a, other volatile organic compounds, or data obtained from a sensor for detecting any combination thereof, the method for odor detection and monitoring according to claim 1.

17. The method for odor detection and monitoring according to claim 1, wherein the measured gas sensor data is obtained from an array sensor having eight or more sensors.

18. (d) A method for detecting and monitoring odor according to claim 1, further comprising the step (40) of displaying odor monitoring results to show at least the measured gas sensor data from at least one sensor, the predicted type of odor, the predicted intensity of the odor, the temperature, and the humidity.

19. The method for odor detection and monitoring according to claim 18, wherein the additional data displayed can be selected from an odor type prediction model, an odor intensity prediction model, a radar chart representing the odor composition, the status of a gas sensor, the location of a gas sensor, local weather data, an odor diffusion pattern, and a history of odor monitoring data.

20. An odor detection and monitoring system comprising an odor database (110), at least one gas sensor (120), a temperature sensor (130), a humidity sensor (140), a processing unit (150), and a user interface unit (160), The odor database (110) stores data used for training odor analysis models and for odor analysis. The gas sensor (120) measures an odor sample diffused in the air within the odor monitoring area, and transmits the measured gas sensor data to the processing unit via communication means. The temperature sensor (130) collects ambient air temperature data when the gas sensor (120) measures an odor sample. The humidity sensor (140) collects ambient humidity data when the gas sensor (120) measures an odor sample. The processing unit (150) performs data processing in order to create an odor analysis model and analyze the odor. In an odor detection and monitoring system in which the user interface unit (160) displays data to the user or receives input from the user, The odor database (110) is divided into at least two datasets, comprising an odor analysis model training dataset and an odor analysis model dataset, wherein the odor analysis model training dataset comprises the type and intensity of odors labeled by at least one panelist, measured gas sensor data from at least one sensor used to measure the same odor samples labeled by the panelist, temperature data and humidity data during the measurement, and the odor analysis model dataset comprises at least one odor analysis model and its parameters. The processing unit (150) uses at least one odor analysis model to process the measured gas sensor data together with temperature data and humidity data by using machine learning techniques. Odor detection and monitoring system.

21. The processing unit (150) (b-1) A step of preprocessing the training dataset for the odor analysis model, (b-2) A step of transforming the preprocessed dataset, (b-3) A step of training at least one odor analysis model, (b-4) A step of evaluating the performance of the trained odor analysis model, (b-5) A step of storing the trained odor analysis model and its parameters. To create an odor analysis model, data processing is performed by having at least the following: The odor detection and monitoring system according to claim 20.

22. The processing unit (150) (c-1) Step (120) of storing measured gas sensor data from at least one gas sensor, (c-2) A step of storing the temperature data and humidity data, (c-3) The step of processing the measured gas sensor data from at least one gas sensor (120) together with temperature data and humidity data, and at least one odor analysis model from the odor analysis model dataset, (c-4) A step of predicting the type and intensity of the odor from the measured data. By having at least, The odor detection and monitoring system according to claim 20, comprising performing data processing to analyze the type and intensity of the odor.

23. The odor detection and monitoring system according to claim 20, wherein the type and intensity of the odors labeled by the panelists are obtained by (1) collecting the odor samples in an odor bag and then analyzing the odors by smelling the odor samples according to the methods of the American Society for Testing and Materials (ASTM) or the Japanese Industrial Standards (JIS), or (2) smelling the odors in the environment using a portable olfactory technique in accordance with the EN13725:2022 standard.

24. The odor detection and monitoring system according to claim 20, wherein the odor analysis model comprises an odor type prediction model and an odor intensity prediction model.

25. The odor detection and monitoring system according to claim 24, wherein the odor type prediction model can be selected from decision trees, random forests, naive Bayes, support vector machines, K-nearest neighbors (KNN), gradient boosting machines, light gradient boosting, extreme gradient boosting, extreme randomized tree models, distributed random forests, generalized linear models, stacked ensembles, deep learning, multilayer perceptrons, recurrent neural networks, long short-term memory networks, generative adversarial networks, and autoencoder deep learning algorithms.

26. The odor detection and monitoring system according to claim 24, wherein the odor type prediction model is an extreme randomization tree model.

27. The odor detection and monitoring system according to claim 24, wherein the odor intensity prediction model can be selected from linear regression, logistic regression, polynomial regression, Lasso regression, Bayesian linear regression, principal component regression, partial least squares regression, and elastic net regression.

28. The odor detection and monitoring system according to claim 24, wherein the odor intensity prediction model is polynomial regression.

29. The odor detection and monitoring system according to claim 20, wherein the processing unit (150) further comprises a filtering unit for filtering the measured gas sensor data.

30. The odor detection and monitoring system according to claim 29, wherein the filtering unit is an exponential filter.

31. The odor detection and monitoring system according to claim 20, wherein the processing unit (150) further comprises a normalization unit that normalizes the measured gas sensor data by using baseline data obtained from clean air measurements.

32. The odor detection and monitoring system according to claim 20, wherein the gas sensor (120) is a metal oxide gas sensor.

33. The gas sensor (120) detects hydrocarbons (C Ｘ H Ｙ ), hydrogen gas (H ２ ), carbon monoxide (CO), alcohol (C Ｘ H Ｙ OH), nitrogen dioxide (NO ２ ), nitric oxide (NO), ozone (O ３ ), ammonia (NH ３ ), sulfur dioxide (SO ２ The odor detection and monitoring system according to claim 20, wherein the sensor is for detecting any of the following: refrigerant R134a, other volatile organic compounds, or combinations thereof.

34. The odor detection and monitoring system according to claim 20, wherein the gas sensor (120) is an array sensor having eight or more sensors.

35. The odor detection and monitoring system according to claim 20, wherein the user interface unit (160) comprises at least a portion for displaying measured gas sensor data from at least one sensor, a portion for displaying predicted odor type data, a portion for displaying predicted odor intensity data, a portion for displaying the temperature sensor data, and a portion for displaying the humidity sensor data.

36. The odor detection and monitoring system according to claim 35, wherein the user interface unit (160) further comprises a data display section for displaying data that can be selected from an odor type prediction model, an odor intensity prediction model, a radar chart representing the odor composition, the status of a gas sensor, the location of a gas sensor, local weather data, a predicted odor diffusion pattern, and an odor monitoring data history.

37. The odor detection and monitoring system according to claim 35, wherein the user interface unit (160) further comprises a portion for displaying the deviation monitoring of the odor analysis model.

38. The odor detection and monitoring system according to claim 35, wherein the user interface unit (160) further comprises a portion for setting and displaying a warning when the odor intensity level exceeds a threshold.

Images