Intelligent sensing data detection method and system

Abstract

A sensing data intelligent detection method and system, the system includes a computer system, provided with a memory, used to store a program set for executing the method and an algorithm program, and a database used to store sensing data obtained from an object to be detected. In the sensing data intelligent detection method, when the system obtains the sensing data, the sensing data is divided into training data and test data, a deep learning algorithm is executed on the training data to obtain features about the object to be detected, a detection model based on deep learning is established, the detection model is tested with the test data, then a machine learning algorithm is executed on the test data that fails the test, the detection model is trained with historical parameter data, the detection model is optimized with a model optimization algorithm, and the optimized control parameters of the object to be detected are obtained.

Description

Technical Field

[0001] The specification discloses a method for detecting sensor data, specifically a method and system for intelligent detection of sensor data that utilizes deep learning and machine learning to build a detection model from the sensor data. Background Technology

[0002] When detecting anomalies in a product, system, or field, a common approach is to use specific sensors to detect data, such as sensing sound or capturing images. Then, analysis tools are used to analyze the information in the sensed data to determine whether the object under test has any abnormal conditions.

[0003] For example, to determine if an object's surface has defects, a camera can be used to photograph the surface. By comparing the image of the object's surface with a sample image, any abnormalities can be identified. Similarly, with motors, a common method is to use a sound sensor to record the sound generated during operation. By comparing this sound with a sample image, it can be determined whether there are any operational abnormalities, serving as a basis for future improvements. Summary of the Invention

[0004] The specification discloses a method and system for intelligent detection of sensing data. The method runs in a computer system. The computer system has a memory that stores the program set and algorithm for executing the intelligent detection method of sensing data. The computer system also has a database that stores the sensing data obtained from an object to be detected.

[0005] In one embodiment, the sensing data is data obtained by sensing the object to be detected through a sensor. In the intelligent detection method of sensing data, the sensing data is first divided into training data and test data. A deep learning algorithm is executed on the training data to obtain features about the object to be detected and to establish a detection model for detecting the object to be detected. Then, the detection model is tested with the test data. For the test data with a failure result, a machine learning algorithm is executed, and the detection model is trained with historical parameter data. The detection model is optimized with a model optimization algorithm to obtain the optimized control parameters of the object to be detected.

[0006] Furthermore, validation data is derived from the training data. During the process of forming a detection model from the training data using a deep learning algorithm in the aforementioned steps, this validation data can be used to validate the detection model, thereby deriving a parameter model used to generate control parameters. These control parameters are the parameters that drive the operation of the object to be detected; generating optimized control parameters is one of the main objectives of the method.

[0007] Furthermore, the parameter model can be further optimized using a model optimization algorithm to generate optimized control parameters, which can then be used to drive the object under test. This process is repeated to obtain sensing data again. The control parameters of the object under test can be optimized using a smart detection method that repeats sensing data.

[0008] Furthermore, in one embodiment, a K-fold cross-validation method can be used to evaluate multiple models generated from multiple deep learning algorithms and machine learning methods using validation data. The evaluation factors include, for example, the accuracy, precision, and recall of each model generated from the multiple models generated by the multiple deep learning algorithms, and then the detection model is selected from them based on the scores. Attached Figure Description

[0009] Figure 1 A system architecture embodiment diagram of the intelligent detection system for sensing data is shown.

[0010] Figure 2 This diagram illustrates a main flowchart of an intelligent detection method for sensor data.

[0011] Figure 3 This illustrates one embodiment of a process for intelligent detection of sensor data during system operation;

[0012] Figure 4 This is the second embodiment of the process for intelligent detection of system operation sensor data;

[0013] Figure 5 A schematic diagram illustrating an operational embodiment of training data, validation data, and test data in an intelligent detection method for sensing data is shown.

[0014] Figure 6 A flowchart illustrating an embodiment of the intelligent detection method for sensor data is shown;

[0015] Figure 7 An optimized example flowchart of the intelligent detection method for sensed data is shown; and

[0016] Figure 8 This diagram illustrates an example flowchart of an intelligent detection method that applies sensor data. Detailed Implementation

[0017] The specification discloses a method and system for intelligent detection of sensor data. One of the main objectives of this method is to utilize deep learning algorithms for initial model training and establishment, and to combine this with machine learning algorithms for subsequent testing and training. This allows for the training of a detection model from sensor data to detect specific objects. For example, the object to be detected can be a product, a system, or a field. Sensors are used to sense specific information about the object, such as images, sounds, and vibrations. The resulting sensor data includes images captured from the object, sound recordings of the object and their converted spectrum, or information about vibrations generated during the object's operation. This sensor data can be used to detect various objects and optimize the control parameters driving the object's operation.

[0018] An embodiment of the intelligent detection system for the sensing data can be found in the following examples. Figure 1 The system architecture example diagram is shown.

[0019] The figure shows an object 10 to be detected. The object operates driven by control parameters. The intelligent detection system based on sensing data can optimize its own control parameters or the control parameters of the system that generates the object by learning the sensing data. For example, the object 10 is sensed by one or more sensors (sensor 101, sensor 202, and sensor 303) to obtain sensing data. If the object 10 is captured to produce an image, sensor 101, sensor 202, and sensor 303 can be a multi-angle camera; if the environmental conditions of a certain field are sensed, sensor 101, sensor 202, and sensor 303 can be sensors located at different positions in the field. The sensing data generated by sensor 101, sensor 202, and sensor 303 can be stored in the sensing data processing host 12, and then converted and preliminarily processed into data in the database 145 of the computer system 14.

[0020] The computer system 14 includes a processor 141, a memory 143, and a database 145. The memory 143 is used to store the program set and algorithm for executing the intelligent detection method of sensing data. The database 145 is used to store the sensing data obtained after measuring the object 10 to be detected by one or more sensors. The intelligent detection method of sensing data is then executed by the processor 141 of the computer system 14.

[0021] Using the above system architecture, Figure 2 The diagram illustrates the main flowchart of the intelligent detection method for sensing data executed by the system.

[0022] The system begins by acquiring sensing data from the object to be detected, as in step S201. Then, as in step S203, the sensing data is divided into training data and testing data. In a specific embodiment, validation data can be derived from the training data. In step S205, a deep-learning algorithm is executed on the training data to obtain features about the object to be detected. These features reflect information about the operation of the object or related systems. A detection model for detecting the object is then built from these features; this is a deep learning-based detection model.

[0023] Next, as in step S207, the detection model obtained in the above steps is tested using the test data selected from the sensing data. The detection result includes the parts that pass the test and the parts that fail the test. Data that passes the test indicates that the detection model meets expectations, meaning that the model parameters obtained from the current detection model are appropriate and do not need to be adjusted. However, if the detection result includes the parts that fail the test, in the intelligent detection method for sensing data, as in step S209, a machine learning algorithm can be executed on the test data that fails, and the detection model obtained in the above steps can be trained with historical parameter data. A parameter model can be obtained by training with historical parameter data, and then the detection model can be optimized with a model optimization algorithm. This is a machine learning-based detection model. As in step S211, the optimized control parameters of the object to be detected can be obtained.

[0024] Machine learning methods encompass a wide range, such as Support Vector Machine, Random Forest, and Simple Bayesian methods. Bayes and deep learning algorithms, etc. The deep learning algorithms mentioned above are a type of machine learning method. They utilize the computing power of the processor in a computer system to transform a large amount of sensor data through multiple processing layers using linear or non-linear transformations. The feature extraction step then extracts features from the data that represent the sensor data characteristics of the object to be detected.

[0025] Common deep learning algorithms employ two methods: Convolutional Neural Networks (CNN) or Recurrent Neural Networks (RNN). CNN can use convolutional operations in its convolutional layers to filter sensed data, gradually extracting features with the computing power of the computer processor to eventually build a model (such as the detection model mentioned above), which can be used to detect the sensed data generated by the system.

[0026] For example, in the method, the system can use a portion (e.g., 90%) of the sensed data obtained from the object to be detected, such as image data, sound data, vibration data, etc., as training data. The training data is used to build a detection model through deep learning algorithms (e.g., CNN, RNN). The remaining portion (e.g., 10%) of the sensed data is used as testing data to test the detection model built by the system.

[0027] Figure 3 This document illustrates a flowchart of an intelligent detection method for system-operated sensing data.

[0028] At the beginning of the process, sensing data can be generated from the object to be detected (301), and sensing data for building a model of intelligent sensing data (303) can be formed. The sensing data or a part thereof (such as training data obtained from the sensing data) is used to execute a deep learning algorithm (305) to build a deep learning-based detection model. Then, the detection model is tested with partial data (such as test data obtained from the sensing data) (307). For the part that passes the test, that is, the process that does not require subsequent parameter optimization, the process ends (309); for the part that fails the test, it is used to build a parameter model (311).

[0029] In the process of establishing the parameter model (311), the data that failed the test is subjected to machine learning again. Historical parameter data is obtained from the historical parameter database (313) in the system. The parameters are trained to establish the parameter model. The model parameters of the detection model are optimized by a model optimization algorithm (315) to form a machine learning-based detection model. Finally, optimized control parameters (317) are generated, which can be re-introduced into the object to be detected to continue generating sensing data (301).

[0030] It should be noted that the control parameters are parameters that drive the operation of the object to be detected, or operating parameters of the system that generates the object to be detected. In particular, in one embodiment, validation data can also be derived from the training data. When the training data is used to form a detection model through a deep learning algorithm, the validation data can be used to validate the detection model to obtain a parameter model for generating the control parameters. Then, the parameter model is optimized by the model optimization algorithm (315), and the optimized control parameters are output.

[0031] Thus, the optimized control parameters (317) drive the object to be detected, and the sensing data (301) is obtained again. The control parameters of the object to be detected are continuously optimized by repeating the intelligent detection method of the sensing data until the system defaults to an expected state.

[0032] Figure 4 Next, another embodiment of the system operation process is shown. This embodiment shows that the continuously generated sensing data (303) stored in the system becomes a historical sensing database (400). This historical sensing database 400 continues to undergo deep learning (305), using the computing power of the computer system to continuously train the data and optimize the detection data. On the other hand, after the detection model is tested, the parameters in the test data can become part of the historical parameter database (313), and machine learning is continuously and repeatedly performed to optimize the parameter model.

[0033] According to one embodiment, when executing the model optimization algorithm, various parameter models can be used, such as one of the following: Genetic Algorithms, Particle Swarm Optimization, Ant Colony Optimization, Simulated Annealing, TabuSearch, Seagull Optimization Algorithm, and Bayesian Optimization. The parameter models used for optimization include Gradient Boosting Decision Tree, Extreme Gradient Boosting, Categorical Boosting, Light GBM, Random Forest, Support Vector Machine, Relevance Vector Machine, and Simple Bayesian Classification. The system may employ Bayesian optimization, K-nearest neighbor (K-nearestneighbor) algorithms, CNN models, or RNN models. In a preferred embodiment, the system may use Bayesian optimization as the model optimization algorithm, a CNN model as the deep learning algorithm, and Extreme Gradient Boosting as the parameter model.

[0034] According to one embodiment, the intelligent detection method for sensing data can employ K-fold cross-validation. This validation method involves dividing the training data into multiple groups (K data sets), with one group serving as the validation data for the model and the other groups (K-1 data sets) serving as the training data. The K groups are then used in rotation for K validation iterations, ultimately yielding the parameters for evaluating the model. Specifically, this K-fold cross-validation method evaluates multiple models generated from various deep learning algorithms using the validation data, and then selects the detection model from among them. One method for evaluating the multiple models involves deriving factors such as accuracy, precision, and recall for each model, including evaluation equations derived from these factors.

[0035] For example, the K-fold cross-validation method, taking 10-fold cross-validation as an example, divides the acquired sensing data into 10 equal parts during computation. The first part of the data can be used as test data for the testing model, and the remaining 9 parts can be used as training data. Then, in the next round of the process, the second part of the data will be used as test data for the detection model, and the remaining 9 parts will also be used as training data. This process is repeated for a total of 10 rounds, achieving an accuracy of 10. The average value is then used to obtain a relatively objective accuracy.

[0036] In one embodiment, "accuracy," "precision," and "recall" can be derived based on the correlation between the predicted results (YES / NO) of each detection model and the actual detection results (YES / NO) obtained from the actual measurement of the object to be detected, and these can be used as the basis for evaluating the detection model.

[0037] For example, data where the model predicts "yes" (YES) and the actual detection result is also "yes" (YES) is set as "TP"; data where the model predicts "yes" (YES) but the actual detection result is "no" (NO) (indicating a prediction error) is set as "FP"; data where the model predicts "no" (NO) but the actual detection result is "yes" (YES) (indicating a prediction error) is set as "FN"; and data where the model predicts "no" (NO) and the actual test result is also "no" (NO) is set as "TN".

[0038] Based on the above settings, "accuracy" represents the proportion of correct predictions made by the model, and the mathematical formula is: "Accuracy = (TP + TN) / N" total In other words, the sum of the data where the model predicted "yes" and the actual test result was also "yes" ("TP") plus the data where the model predicted "no" and the actual test result was also "no" ("TN") divided by the total number of data (N). total ).

[0039] The "accuracy" mentioned above is one of the main objectives of the entire system to optimize the detection model. The mathematical formula is: "Precision = TP / (TP + FP)", which is the proportion of the number of actual detection results that are also predicted as "yes" by the model ("TP") to the total number of predictions that are "yes" by the model (i.e., "TP" + "FP").

[0040] The "recall rate" represents the model's error rate, and its mathematical formula is: "Recall = TP / (TP + FN)", which means the proportion of data that the model predicted as "yes" and the actual detection result was also "yes" ("TP") to all data that were actually detected as "yes" (i.e., "TP" + "FN").

[0041] Furthermore, the calculation results of "accuracy", "precision" and "recall" can be used for other operations (such as F1 Score = 2 / ((1 / Precision) + (1 / Recall))) as a basis for evaluating the quality of the model.

[0042] Figure 5 This diagram illustrates an implementation example of training data, validation data, and test data in an intelligent detection method for sensing data. The following is accompanied by... Figure 6 The flowchart of an embodiment of the intelligent detection method for sensing data is shown below for illustration.

[0043] The system database 50 contains sensing data and parameter data (step S601). The sensing data can be mainly divided into training data 501 and test data 505, and a portion of the training data 501 can be used as validation data 503 (step S603). Based on the training data 501, a model (52) is trained using a deep learning algorithm to obtain a detection model (step S605). The model is tested using the test data 505 (56) (step S607) to obtain a parameter model used to generate control parameters. At this time, the control parameters can be generated by machine learning (step S609). On the other hand, the validation data 503 is used to validate the model and obtain a parameter model (step S611).

[0044] According to one embodiment, the parameter model can be optimized by a model optimization algorithm (53) (step S613), and the parameter model can be tested with test data (step S615) to form a detection model (54). By repeating the above process, multiple detection models can be obtained by introducing different deep learning algorithms and repeatedly optimizing them with training data 501, validation data 503 and test data 505 to generate control parameters (step S617). Then, by comparing the models 57, the model suitable for the system is determined according to the requirements (58), and the intelligent detection process of the sensing data is completed (59).

[0045] The methods for the above optimization model can be referred to Figure 7 The flowchart shown is an optimized example of the intelligent detection method for sensing data.

[0046] According to the above embodiments, in the process of optimizing a model using a model optimization algorithm to define the optimal detection model, historical parameter data can be imported (step S701), and a parameter model can be established by training this data with a machine learning algorithm (step S703). Then, control parameters are derived from this parameter model. Multiple models generated from multiple deep learning algorithms are tested one by one using K-fold cross-validation (step S705) with validation data (step S707). The most suitable detection model is then evaluated (e.g., using factors such as "accuracy", "precision", and "recall" as described above) and selected (step S709). Afterwards, the model parameter optimization algorithm is executed on the selected model (step S711) to obtain the optimized detection model (step S713). The detection model can be used to generate the optimal control parameters.

[0047] Here is an example for reference. Figure 8 The flowchart shown illustrates an intelligent method for building a detection model for a motor.

[0048] The vibration sound of the motor is obtained through a sound sensor (step S801), the sound is converted into a spectrum, and data is established in the sensing database (step S803). Then, this sensing data (such as the training data obtained therein) is trained using a deep learning algorithm. Sound features related to the motor's operating status are extracted from the data, and the correlation between various audio frequencies and whether the motor is operating abnormally or not, as well as its control parameters, is obtained. A detection model for detecting the motor is then established (step S805). Afterward, the system uses verification data obtained from the sensing data to verify the detection model trained in the above steps (step S807), and the detection model is determined after verification (step S809).

[0049] The system uses this detection model to generate control parameters to drive the motor. At the same time, it records sound through sensors to form audio sensing data. Then, it detects this sensing data (such as the imaged spectrum) (step S811) and determines whether the test is passed based on the sensing data (step S813). If the test is passed, it means that the control parameters determined by the current detection model to drive the motor meet the system requirements, and the process ends (step S815).

[0050] If the test fails (No), it means that the parameters need to be continuously optimized using the parameter model (step S817), and then the motor is driven to operate using the generated control parameters (step S819). The vibration sound is continuously obtained by the sensor, and after being converted into a spectrum, the detection model detects the data to determine whether the test is passed. The above steps are repeated to continuously optimize the parameter model to obtain the optimal control parameters.

[0051] Furthermore, according to another application embodiment, the disclosed intelligent detection method for sensing data is applied to the detection of a screen to demonstrate parameters that drive the screen to display content. As described above, a large amount of screen image data can be obtained first using an image sensor, and features in the image data can be obtained using a deep learning algorithm to establish the correlation between the image and the screen display parameters. A detection model for the screen can be established, including determining the detection model by verifying with verification data, then using this detection model to generate display parameters, and then generating screen image data for testing, and executing subsequent machine learning algorithms to optimize the detection model to obtain optimized display parameters for the screen.

[0052] In summary, the intelligent detection method and system for sensing data described in the above embodiments are characterized by the following: When building a model using training data, in addition to using a portion of the sensing data as training data, a portion is also used as testing data. Furthermore, a portion of the training data is used as validation data to verify the detection model generated by the system. This is an internal optimization step. Moreover, during model training, various deep learning and machine learning algorithms can be employed to generate multiple models. These models can be optimized, and finally, a comparison and evaluation of the models is performed before selecting one for practical application.

[0053] However, the above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Therefore, all equivalent structural changes made based on the description and illustrations of the present invention are similarly included within the scope of the present invention and are hereby stated.

Claims

1. A method for intelligent detection of sensor data, characterized in that, The method comprises: obtaining sensing data of a to-be-detected object, and dividing the sensing data into training data and test data, the sensing data being obtained by one or more sensors sensing the to-be-detected object, the sensing data obtained by the one or more sensors sensing the to-be-detected object at least including image data, sound data, or vibration data, the to-be-detected object at least including a product, a system, or a field; performing a plurality of deep learning algorithms on the training data, obtaining features of the to-be-detected object from the training data, obtaining validation data from the training data, evaluating a plurality of models generated from the plurality of deep learning algorithms by using the validation data in a K-fold cross-validation method, and selecting a detection model for detecting the to-be-detected object from the plurality of models; testing the detection model by using the test data; and for test data that fails the test, integrating historical parameter data, performing a machine learning algorithm, training the historical parameter data and the test data that fails the test integrated into the historical parameter data by using the machine learning algorithm, establishing a parameter model, obtaining control parameters by using the parameter model, training the detection model by using the historical parameter data, and optimizing the detection model by using a model optimization algorithm to obtain optimized control parameters of the to-be-detected object. The to-be-detected object is driven to operate by using the control parameters, and the control parameters can be optimized by learning the sensing data of the to-be-detected object. The control parameters are parameters for driving the to-be-detected object to operate or operation parameters of a system generating the to-be-detected object. The correlation between the predicted results of the detection model and the actual detection results obtained by actually measuring the to-be-detected object at least includes an accuracy, a precision, or a recall rate to evaluate the detection model.

2. The method of claim 1, wherein the sensing data intelligent detection method is characterized by, The parameter model is further optimized by using the model optimization algorithm, and the optimized control parameters are generated.

3. The method of claim 2, wherein the sensing data intelligent detection method is characterized by, The optimized control parameters are used to drive the to-be-detected object, and sensing data is obtained again. The control parameters of the to-be-detected object are optimized by repeating the sensing data intelligent detection method.

4. The method of claim 1, wherein The basis for evaluating the detection model includes the accuracy, the precision, and the recall rate of each model in the plurality of models generated from the plurality of deep learning algorithms.

5. A system for intelligent detection of sensed data, the system comprising: The system comprises: a computer system provided with a memory storing a set of programs and algorithms for executing a sensing data intelligent detection method, and a database storing sensing data obtained from a to-be-detected object, the sensing data being obtained by one or more sensors sensing the to-be-detected object, the sensing data obtained by the one or more sensors sensing the to-be-detected object at least including image data, sound data, or vibration data, the to-be-detected object at least including a product, a system, or a field; wherein a processor of the computer system executes the sensing data intelligent detection method, including: obtaining the sensing data, and dividing the sensing data into training data and test data; performing a plurality of deep learning algorithms on the training data, obtaining features of the object to be detected from the training data, deriving validation data from the training data, evaluating a plurality of models generated from the plurality of deep learning algorithms using the validation data in a K-fold cross-validation method, and selecting a detection model from the plurality of models for detecting the object to be detected; testing the detection model using the test data; for test data that fails the test, integrating historical parameter data, performing a machine learning algorithm, training the historical parameter data and the test data that fails the test using the machine learning algorithm, establishing a parameter model, deriving control parameters using the parameter model, training the detection model using historical parameter data, optimizing the detection model using a model optimization algorithm, and deriving optimized control parameters of the object to be detected; wherein the object to be detected is driven to operate using the control parameters, and the control parameters can be optimized by learning sensing data of the object to be detected, the control parameters being parameters for driving the object to be detected to operate or operation parameters of a system generating the object to be detected; wherein the accuracy, precision, or recall between the predicted results of the detection model and the actual detection results obtained by actually measuring the object to be detected can be used to evaluate the detection model.

6. The intelligent sensing data detection system of claim 5, wherein, The parameter model is further optimized using the model optimization algorithm, and the optimized control parameters are generated, and the object to be detected is continuously driven, sensing data is obtained again, and the control parameters of the object to be detected are optimized by repeating the sensing data intelligent detection method.

7. The intelligent sensing data detection system of claim 5 or 6, wherein, The plurality of models generated from the plurality of deep learning algorithms are evaluated using the validation data in a K-fold cross-validation method, and the detection model is selected from the plurality of models; wherein the basis for evaluating the detection model includes the accuracy, precision, and recall of each model from the plurality of models generated from the plurality of deep learning algorithms.

Images