An underwater robot fault identification method and system
By calculating the correlation and power relationship of the operating data of various systems of the underwater robot, abnormal data can be identified, solving the problem of distinguishing between underwater robot faults and sensor faults, and improving the accuracy and reliability of fault detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SU ZHOU SHI HANG ZHI NENG KE JI YOU XIAN GONG SI
- Filing Date
- 2024-07-05
- Publication Date
- 2026-06-23
Smart Images
Figure CN121300482B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data processing technology, and specifically to a method and system for fault identification of underwater robots. Background Technology
[0002] Underwater robots are designed to perform various tasks in fields such as deep-sea exploration, marine scientific research, marine resource exploration, subsea pipeline repair, and underwater rescue. Because malfunctions in underwater robots can prevent them from completing their intended tasks, thus affecting the progress of research, exploration, or rescue operations, timely fault detection and repair can maximize mission completion rates. Therefore, fault detection for underwater robots is crucial.
[0003] Currently, the determination of whether an underwater robot is malfunctioning is based on data transmitted from sensors on the robot's body. However, because these sensors can also malfunction, when abnormal operational data appears, it's impossible to determine whether the anomaly is caused by a malfunction in the robot itself or a faulty sensor. This could lead to a misidentification of a sensor malfunction as a robot malfunction. Summary of the Invention
[0004] This invention provides a method and system for fault identification of underwater robots to solve existing problems.
[0005] The underwater robot fault identification method and system of the present invention adopts the following technical solution:
[0006] This invention proposes a fault identification method for underwater robots, which includes the following steps:
[0007] The underwater robot uses sensors inside to acquire a sequence of operational data in each dimension of each system during operation; each piece of operational data corresponds to a numerical value, a time, and the number of times the operation is performed.
[0008] Based on the number of times each running data is run, each running data sequence is divided into several data segments; based on the values within each data segment, each data segment is further divided into several new data segments; based on the position of each new data segment within each data segment, the position of each running data within each new data segment, and the number of times each running data is run, several data combinations are obtained, each data combination containing running data of two different dimensions; based on the time differences of running data of different dimensions in each data combination and the numerical differences of running data of the same dimension in different data combinations, the probability that the value of running data of each dimension is an exponential power of the values of running data of other dimensions is obtained, thereby obtaining the correlation dimension of each dimension, the correlation relationship between each dimension and the correlation dimension, and the exponential relationship between the value of running data of each dimension and the value of running data of the correlation dimension.
[0009] Based on the numerical values of each operational data point, abnormal data in each dimension are obtained; based on the time of each abnormal data point in each dimension and the time of each abnormal data point in the correlation dimension, the comparison data of each abnormal data point in each dimension in each correlation dimension is obtained; based on the abnormal data and comparison data in each dimension, the operational data to the left of the abnormal data and the operational data to the left of the comparison data in the operational data sequence of each dimension, the correlation between each dimension and the correlation dimension, and the power relationship between the numerical values of the operational data in each dimension and the numerical values of the operational data in the correlation dimension, fault data is obtained, and anomaly detection is completed.
[0010] Furthermore, the specific steps for dividing each running data sequence into several data segments based on the number of times each running data is executed are as follows:
[0011] Taking the underwater robot's propulsion system as the target system, in the operational data sequence of the a-th dimension within the target system, operational data with the same number of runs are grouped into one data segment, and the operational data of the a-th dimension within the target system are divided into several data segments.
[0012] Furthermore, the specific steps for dividing each data segment into several new data segments based on the values within the data segment are as follows:
[0013] The mode number of the values of the running data in any data segment of the running data sequence in the a-th dimension of the target system is recorded as the normal value of that data segment in the a-th dimension of the target system when the power is fixed.
[0014] Remove the running data in the a-th dimension of the target system from the data segment where the value is the normal value of the data segment in the a-th dimension of the target system when the power is fixed, and obtain the continuous data interval in the data segment of the running data sequence in the a-th dimension of the target system.
[0015] The continuous data interval within each data segment of the operational data sequence in the a-th dimension of the target system is denoted as a new data segment in each data segment of the a-th dimension of the target system.
[0016] The new data segment within each data segment of the running data sequence of the a-th dimension in the target system is denoted as the new data segment in the a-th running data sequence; each new data segment corresponds to a running number and a position.
[0017] Furthermore, the specific steps for obtaining several data combinations based on the position of each new data segment within each data segment, the position of each running data within each new data segment, and the number of times each running data is executed are as follows:
[0018] In the a-th and b-th running data sequences, new data segments that have the same number of runs and are in the same position are designated as corresponding data segments.
[0019] The i-th running data in a new data segment of the a-th running data sequence and the i-th running data in the corresponding data segment are corresponding data;
[0020] The running data in the new data segment of the a-th running data sequence is combined with its corresponding data to form a data combination, resulting in several data combinations. The data combinations are then sorted according to the time of the running data in the a-th dimension from smallest to largest.
[0021] Furthermore, the specific steps for determining the power probability of the value of the running data in each dimension being the value of the running data in other dimensions, based on the time differences of the running data in different dimensions within each data combination and the numerical differences of the running data in the same dimension across different data combinations, are as follows:
[0022]
[0023]
[0024] In the formula, R b,a,k R represents the probability that the value of the b-th dimension of the running data in the target system is a power of k of the value of the a-th dimension of the running data. a,b,k This represents the probability that the value of the running data in the a-th dimension of the target system is a power of k of the value of the running data in the b-th dimension. This represents the value of the running data in the a-th dimension of the (i+1)-th data combination. This represents the value of the b-th dimension of the running data in the (i+1)-th data combination. Let k represent the k-th power of the value of the data in the a-th dimension of the (i+1)-th data combination. Let k represent the k-th power of the value of the data in the a-th dimension of the i-th data combination. Let k represent the value of the b-th dimension of the running data in the (i+1)-th data combination raised to the power of k. Let k represent the k-th power of the value of the b-th dimension of the running data in the i-th data combination. This represents the numerical value of the running data in the a-th dimension of the i-th data combination. This represents the value of the b-th dimension of the running data in the i-th data combination. This represents the mean of the absolute values of the differences in time between two running data points across all data combinations. This represents the variance of the ratio of the k-th power of the value of the running data in the a-th dimension of all data combinations to the k-th power of the value of the running data in the a-th dimension of the previous data combination, to the ratio of the k-th power of the value of the running data in the b-th dimension of all data combinations to the k-th power of the value of the running data in the b-th dimension of the previous data combination. Exp() represents the variance of the ratio of the k-th power of the value of the b-th dimension's running data in all data combinations to the k-th power of the value of the b-th dimension's running data in the previous data combination, to the ratio of the b-th dimension's running data in all data combinations to the difference between the b-th dimension's running data in all data combinations and the difference between the b-th dimension's running data in the previous data combination. k represents the exponent, K represents the preset maximum value of the exponent, and exp() is an exponential function with the natural constant as the base.
[0025] Furthermore, the specific steps for obtaining the correlation dimension of each dimension, the correlation between each dimension and the correlation dimension, and the power relationship between the numerical values of the operational data of each dimension and the numerical values of the operational data of the correlation dimension are as follows:
[0026] Get the maximum R a,b,k The value of R and the maximum R b,a,k The value;
[0027] If the maximum R a,b,k The value is greater than the maximum R b,a,k The value of R will be the maximum. a,b,k The value β is denoted as the correlation between the operational data of the a-th dimension and the b-th dimension within the target system. a,b , will the maximum R a,b,k The corresponding k is denoted as R. a,b ;
[0028] If the maximum R a,b,k The value is less than the maximum R b,a,k The value of R will be the maximum. b,a,k The value β is denoted as the correlation between the operational data of the a-th dimension and the b-th dimension within the target system. a,b , will the maximum R b,a,k corresponding Let it be R a,b ;
[0029] If a preset correlation threshold T1 is set, and β a,b ≥T1, where the a-th and b-th dimensions within the target system are correlated dimensions, β a,b R represents the correlation between the a-th dimension and the b-th dimension within the target system. a,b This represents the power relationship between the values of the operational data in the a-th dimension and the values of the operational data in the b-th dimension within the target system.
[0030] Furthermore, the specific steps for obtaining abnormal data in each dimension based on the value of each running data are as follows:
[0031] The normal range [T1, T2] of the operating data of the a-th dimension in the target system is preset when the robot is running normally, and the normal range [T3, T4] of the operating data of the j-th related dimension of the a-th dimension in the target system is preset. T1 represents the minimum value of the operating data of the a-th dimension in the target system, T2 represents the maximum value of the operating data of the a-th dimension in the target system, T3 represents the minimum value of the operating data of the j-th related dimension of the a-th dimension in the target system, and T4 represents the maximum value of the operating data of the j-th related dimension of the a-th dimension in the target system.
[0032] The running data in the running data sequence of the a-th dimension within the target system that is not within the interval [T1,T2] is denoted as abnormal data in the a-th dimension; the running data in the running data sequence of the j-th correlation dimension of the a-th dimension within the target system that is not within the interval [T3,T4] is denoted as abnormal data in the j-th correlation dimension of the a-th dimension.
[0033] Furthermore, the specific steps for obtaining the comparison data of each anomalous data point in each dimension for each correlation dimension based on the time of each anomalous data point in each dimension and the time of each anomalous data point in the correlation dimension are as follows:
[0034] The outlier in the j-th correlation dimension of the a-th dimension that is closest in time to the v-th outlier in the a-th dimension is denoted as the reference data of the v-th outlier in the j-th correlation dimension.
[0035] Furthermore, the specific steps for obtaining fault anomaly data based on the abnormal data and control data of each dimension, the running data to the left of the abnormal data and the running data to the left of the control data in the running data sequence of each dimension, the correlation between each dimension and the correlation dimension, and the power relationship between the value of the running data of each dimension and the value of the running data of the correlation dimension are as follows:
[0036]
[0037] In the formula, ε a,v This represents the probability that the v-th outlier in the a-th dimension is caused by a fault, where n is the number of outliers. a β represents the number of relevance dimensions in the target system at the a-th dimension. 1,a,j This represents the correlation between the a-th dimension and the j-th correlation dimension within the target system, where a 1,v This represents the value of the v-th outlier within the a-th dimension, where a 2,v-1 This represents the value of the running data to the left of the v-th abnormal data point within the running data sequence of the a-th dimension of the target system. R represents the value of the comparison data of the v-th outlier in the a-th dimension in the j-th relevance dimension. 1,a,b The value of the power, R represents the value of the running data to the left of the control data in the j-th correlation dimension of the running data sequence in the a-th correlation dimension within the target system, specifically the v-th outlier in the a-th correlation dimension. 1,a,b The value of the power, R 1,a,b This represents the power relationship between the numerical values of the a-th dimension of the target system and the numerical values of the j-th correlation dimension of the operating data. sigmoid() represents the normalization function, and exp() is an exponential function with the natural constant as the base.
[0038] Preset probability threshold T ε When ε a,v ≥T ε At that time, the vth abnormal data in the ath dimension of the target system is the fault data.
[0039] The present invention also proposes an underwater robot fault identification system, including a memory, a processor, and a computer program stored in the memory and running on the processor. The processor executes the computer program stored in the memory to implement the steps of the aforementioned underwater robot fault identification method.
[0040] The beneficial effects of the technical solution of the present invention are as follows: It acquires the operational data sequence for each dimension within each system. When obtaining the data segment of each operational data sequence, it obtains the corresponding data segment for each data segment based on the number of runs within each data segment. This reduces the influence of data segments caused by sensor failures and robot failures on the corresponding data segments of other data segments, laying the groundwork for calculating the power probability of the operational data value of each dimension being the operational data value of other dimensions. When calculating the probability that abnormal data in each dimension is caused by a failure, it reduces the randomness of the calculation results based on the values of abnormal data in all related dimensions of each dimension. Furthermore, it makes the probability that abnormal data caused by sensor failures is caused by a failure far less than the probability that abnormal data caused by robot failures is caused by a failure, resulting in more accurate anomaly detection results. Attached Figure Description
[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0042] Figure 1 This is a flowchart illustrating the steps of an underwater robot fault identification method according to the present invention. Detailed Implementation
[0043] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an underwater robot fault identification method and system proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.
[0044] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0045] The following description, in conjunction with the accompanying drawings, details the specific scheme of the underwater robot fault identification method and system provided by the present invention.
[0046] Please see Figure 1 The diagram illustrates a flowchart of a fault identification method for an underwater robot according to an embodiment of the present invention. The method includes the following steps:
[0047] Step S001: Use the sensors inside the underwater robot to acquire the sequence of operational data in each dimension of each system during the robot's operation.
[0048] The purpose of this embodiment is to determine whether the underwater robot has malfunctioned by using data from its operation. Therefore, data from the underwater robot's operation is acquired.
[0049] It's important to note that underwater robots are often broken down into numerous systems during product design and manufacturing, such as propulsion, lighting, camera, sonar, robotic arm, and power systems. Fault monitoring of an underwater robot essentially involves monitoring each of these systems, and each system has various types of sensors that monitor its operational status from different perspectives. Therefore, the goal is to acquire operational data for each system within each dimension during the underwater robot's operation through these sensors.
[0050] Specifically, using sensors that acquire operational data every M1 seconds, this embodiment, taking the propulsion system as an example, uses five types of underwater robot operational data—propulsion motor current data, propulsion motor voltage data, propulsion motor temperature data, propulsion motor speed data, and propulsion motor torque data—as five dimensions of operational data. The system acquires operational data for each dimension during every run of the underwater robot within the acquisition time M, resulting in an operational data sequence for each dimension. Each piece of operational data in the sequence corresponds to a numerical value, a time period, and a number of runs. The number of runs corresponds to which operational data was acquired during which run of the underwater robot within the acquisition time. The time corresponding to the operational data represents the acquisition time of the operational data. This embodiment presets the acquisition time M=1 and the acquisition interval M1=1 for illustrative purposes; other values can be used in other implementations.
[0051] Step S002: Divide each running data sequence into several data segments based on the number of times each running data is run; divide each data segment into several new data segments based on the values within each data segment; obtain several data combinations based on the position of each new data segment within each data segment, the position of each running data within each new data segment, and the number of times each running data is run; based on the time differences of running data in different dimensions within each data combination and the numerical differences of running data in the same dimension within different data combinations, obtain the power probability that the value of running data in each dimension is the power of the values of running data in other dimensions, and thus obtain the correlation dimension of each dimension, the correlation relationship between each dimension and the correlation dimension, and the power relationship between the value of running data in each dimension and the value of running data in the correlation dimension.
[0052] It's important to note that due to the structural complexity of underwater robots, a malfunction in one part can lead to malfunctions in other parts. Specifically, if the monitoring component corresponding to one dimension within a robot's system fails, it can cause abnormal changes in the operational data of other dimensions within that system. In other words, the operational data of each dimension within each system is correlated with the operational data of other dimensions within that system. However, a sensor malfunction will not cause anomalies in multiple operational data points within the same system. Therefore, when monitoring the robot's operational status using sensor data, the determination of whether a malfunction has occurred is based on the values of the operational data for each dimension within each system and the correlation between the operational data of each dimension within each system and the operational data of other dimensions when a malfunction occurs.
[0053] It should be further noted that when the robot malfunctions, the stronger the correlation between the operational data of each dimension within each system and the operational data of other dimensions within that system, the stronger the correlation between the operational data of the two dimensions mentioned above when the robot is operating normally. Therefore, by measuring the correlation between the operational data of each dimension and the operational data of other dimensions when the robot is operating normally, we can quantify the correlation between the operational data of each dimension and the operational data of other dimensions when the robot malfunctions.
[0054] It's important to further clarify that the correlation between the operational data of each dimension within each system and the operational data of other dimensions within that system during normal robot operation refers to the correlation between the values of the operational data of each dimension within each system and the values of the operational data of other dimensions within that system when the robot's operating power changes. When the robot's power changes during normal operation, the closer the value of the operational data of one dimension within a system is to the power of the value of the operational data of other dimensions within that system, the stronger the correlation between these two dimensions during normal robot operation. Therefore, based on the correlation between the operational data of each dimension within each system and the operational data of other dimensions within that system during normal robot operation, and considering that the value of the operational data of each dimension within each system changes accordingly, we can obtain the data segment where the power changes based on the value of each operational data point within the operational data sequence of each dimension within each system during normal robot operation, thus obtaining the correlation between the operational data of each dimension within each system and the operational data of other dimensions within that system.
[0055] It should be further noted that the probability of sensor or robot malfunction is low during robot operation; that is, the operational data sequence for each dimension within each system mainly consists of data from when the robot is operating normally. Therefore, by using the operational data sequence for each dimension within each system as the data sequence from when the robot is operating normally, the correlation between the operational data for each dimension within each system and the operational data for other dimensions within that system can be obtained.
[0056] Specifically, the underwater robot's propulsion system is taken as the target system. In the operational data sequence of the *a*-th dimension within the target system, operational data with the same number of runs are grouped into a single data segment, thus dividing the operational data of the *a*-th dimension within the target system into several data segments. Each data segment corresponds to one number of runs. The mode of the numerical value of the operational data within any data segment of the operational data sequence of the *a*-th dimension within the target system is denoted as the normal value of that data segment within the *a*-th dimension of the target system when the power is fixed. Operational data within any data segment of the operational data sequence of the *a*-th dimension within the target system whose values are the normal values of that data segment within the *a*-th dimension of the target system when the power is fixed are removed, resulting in continuous data intervals within that data segment of the operational data sequence of the *a*-th dimension within the target system. Each continuous data interval within each data segment of the operational data sequence of the *a*-th dimension within the target system is denoted as a new data segment within each data segment of the *a*-th dimension within the target system. Each data segment of the operational data sequence of the *a*-th dimension within the target system is then divided into several new data segments. A new data segment within each data segment of the operational data sequence in the a-th dimension of the target system is denoted as a new data segment in the a-th operational data sequence. Each new data segment corresponds to a number of runs and a position. The number of runs for a new data segment refers to the number of times the operational data within that new data segment was acquired, and the position of the new data segment refers to which new data segment it is within a set of data segments.
[0057] The above operation is performed on the running data sequence of each dimension within the target system to obtain new data segments in several running data sequences.
[0058] Furthermore, within the target system, in the *a*-th and *b*-th running data sequences, new data segments with the same number of runs and the same position are considered corresponding data segments. The new data segments and corresponding data segments in the *a*-th running data sequence form a data segment combination. If the new data segments in the *a*-th and *b*-th running data sequences with the same number of runs are different from those in the *b*-th running data sequence, then some new data segments do not have corresponding data segments. The *i*-th running data in a new data segment of the *a*-th running data sequence and the *i*-th running data in the corresponding data segment are considered corresponding data. If the number of running data in a new data segment of the *a*-th running data sequence is different from the number of running data in the corresponding data segment, the portion with more running data does not have corresponding data. The running data in the new data segment of the *a*-th running data sequence within the target system and its corresponding data form a data combination. The data combinations are then sorted according to the time of the running data in the *a*-th dimension from smallest to largest. If a running data in the a-th or b-th running data sequence does not have a corresponding data, then this running data has no data combination.
[0059] Furthermore, based on the value and time of each data combination for each running data, the probability that the value of the running data in the b-th dimension of the target system is a power of k is obtained. The specific calculation formula is as follows:
[0060]
[0061]
[0062] In the formula, R b,a,k R represents the probability that the value of the b-th dimension of the running data in the target system is a power of k of the value of the a-th dimension of the running data. a,b,k This represents the probability that the value of the running data in the a-th dimension of the target system is a power of k of the value of the running data in the b-th dimension. This represents the value of the running data in the a-th dimension of the (i+1)-th data combination. This represents the value of the b-th dimension of the running data in the (i+1)-th data combination. Let k represent the k-th power of the value of the data in the a-th dimension of the (i+1)-th data combination. Let k represent the k-th power of the value of the data in the a-th dimension of the i-th data combination. Let k represent the value of the b-th dimension of the running data in the (i+1)-th data combination raised to the power of k. Let k represent the k-th power of the value of the b-th dimension of the running data in the i-th data combination. This represents the numerical value of the running data in the a-th dimension of the i-th data combination. This represents the value of the b-th dimension of the running data in the i-th data combination. This represents the mean of the absolute values of the differences in time between two running data points across all data combinations. This represents the variance of the ratio of the k-th power of the value of the running data in the a-th dimension of all data combinations to the k-th power of the value of the running data in the a-th dimension of the previous data combination, to the ratio of the k-th power of the value of the running data in the b-th dimension of all data combinations to the k-th power of the value of the running data in the b-th dimension of the previous data combination. This represents the variance of the ratio of the k-th power of the value of the b-th dimension's running data in all data combinations to the k-th power of the value of the b-th dimension's running data in the previous data combination, to the ratio of the b-th dimension's running data in all data combinations to the difference between the b-th dimension's running data in all data combinations and the difference between the b-th dimension's running data in the previous data combination. k represents the exponent, and K represents the preset maximum exponent. In this embodiment, the preset maximum power K = 18, which is used as an example for description. Other values can be set in other embodiments. exp() is an exponential function with the natural constant as the base. In this embodiment, the exp(-x) model is used to present the inverse proportional relationship and normalization processing. x is the input of the model. Implementers can set the inverse proportional function and normalization function according to the actual situation.
[0063] It should be noted that, This indicates the correlation between the changes in the operational data of the a-th dimension and the operational data of the b-th dimension within the target system. The smaller the value, the more synchronized the changes in the operational data of the a-th dimension and the operational data of the b-th dimension within the target system are during normal robot operation, that is, the stronger the linear relationship between the operational data of the a-th dimension and the operational data of the b-th dimension in the target system. This indicates the consistency among all data combinations in that the value of the b-th running data is the k-th power of the value of the a-th running data. The smaller the value, the greater the probability that the value of the b-th dimension running data in the target system is the k-th power of the value of the a-th dimension running data.
[0064] Furthermore, we obtain the maximum R. a,b,k The value of R and the maximum R b,a,k The value of R. If the maximum R a,b,k The value is greater than the maximum R b,a,k The value of R will be the maximum. a,b,k The value of β is denoted as the correlation β between the operational data of the a-th dimension and the b-th dimension within the target system. a,b Let k at this moment be denoted as R. a,b If the maximum R a,b,k The value is less than the maximum R b,a,k The value of R will be the maximum. b,a,k The value of β is denoted as the correlation β between the operational data of the a-th dimension and the b-th dimension within the target system.a,b , at this time Let it be R a,b .
[0065] Furthermore, a preset correlation threshold T1 is set; if β a,b ≥T1, where the a-th and b-th dimensions within the target system are correlated dimensions, β a,b R represents the correlation between the a-th dimension and the b-th dimension within the target system. a,b This represents the power relationship between the values of the operational data in the a-th dimension and the values of the operational data in the b-th dimension within the target system. In this embodiment, the preset correlation threshold T1 = 0.6 is used as an example for description. Other values can be set in other embodiments.
[0066] Thus, we obtain the correlation dimension of each dimension within each system, the correlation between each dimension and the correlation dimension, and the power relationship between the numerical values of the operational data of each dimension and the numerical values of the operational data of the correlation dimension.
[0067] Step S003: Based on the value of each running data, obtain the abnormal data in each dimension; based on the time of each abnormal data in each dimension and the time of each abnormal data in the correlation dimension, obtain the reference data of each abnormal data in each dimension in each correlation dimension; based on the abnormal data and reference data in each dimension, the running data to the left of the abnormal data and the running data to the left of the reference data in the running data sequence of each dimension, the correlation between each dimension and the correlation dimension, and the power relationship between the value of the running data in each dimension and the value of the running data in the correlation dimension, obtain the fault data and complete the anomaly detection.
[0068] It should be noted that during normal operation, the operational data of the underwater robot in each dimension within each system fluctuates within a certain range. Therefore, based on the value of each data point in the operational data sequence of each dimension within each system, abnormal data in the operational data sequence of each dimension within each system is obtained. By analyzing the correlation between the abnormal data in the operational data sequence of each dimension within each system and other dimensions within each system, the probability that each abnormal data point is caused by a fault is determined. Further, the abnormal data caused by faults are identified, thus completing fault identification.
[0069] Specifically, the normal range [T1, T2] of the operating data in the a-th dimension of the target system is preset when the robot is running normally, and the normal range [T3, T4] of the operating data in the j-th related dimension of the a-th dimension of the target system is preset. In this embodiment, the minimum value T1 = 5 and the maximum value T2 = 10 of the operating data in the a-th dimension are preset, and the minimum value T3 = 8 and the maximum value T4 = 16 of the operating data in the j-th related dimension of the a-th dimension are preset. This is used as an example for description, and other values can be set in other embodiments.
[0070] Furthermore, operational data in the operational data sequence of the a-th dimension within the target system that is not within the interval [T1, T2] are denoted as abnormal data in the a-th dimension. Operational data in the operational data sequence of the j-th correlation dimension within the a-th dimension that is not within the interval [T3, T4] are denoted as abnormal data in the j-th correlation dimension within the a-th dimension. The abnormal data in the j-th correlation dimension within the a-th dimension that is closest in time to the v-th abnormal data in the a-th dimension is denoted as the reference data for the v-th abnormal data in the j-th correlation dimension. Based on the value of each abnormal data in the a-th dimension and the value of the reference data for each abnormal data in each correlation dimension, combined with the correlation between the a-th dimension and each correlation dimension within the target system, and the power relationship between the values of the operational data in the a-th dimension and the values of the operational data in the j-th correlation dimension, the probability that each abnormal data in each dimension within the target system is caused by a fault is obtained. The specific calculation formula is as follows:
[0071]
[0072] In the formula, ε a,v This represents the probability that the v-th outlier in the a-th dimension is caused by a fault, where n is the number of outliers. a β represents the number of relevance dimensions in the target system at the a-th dimension. 1,a,j This represents the correlation between the a-th dimension and the j-th correlation dimension within the target system, where a 1,v The value of the vth outlier in the a-th dimension, a 2,v-1 This represents the value of the running data to the left of the v-th abnormal data point within the running data sequence of the a-th dimension of the target system. R represents the value of the comparison data of the v-th outlier in the a-th dimension in the j-th relevance dimension. 1,a,b The value of the power, R represents the value of the running data to the left of the control data in the j-th correlation dimension of the running data sequence in the a-th correlation dimension within the target system, specifically the v-th outlier in the a-th correlation dimension. 1,a,b The value of the power, R 1,a,bThis represents the power relationship between the numerical values of the operating data in the a-th dimension and the numerical values of the operating data in the j-th correlation dimension within the target system. sigmoid() represents the normalization function, and exp() is an exponential function with the natural constant as the base. In this embodiment, the exp(-x) model is used to present the inverse proportional relationship, where x is the input of the model. The implementer can set the inverse proportional function according to the actual situation.
[0073] It should be noted that, It reflects the probability that the vth abnormal data in the a-th dimension is caused by a fault, based on the abnormal data in the j-th relevance dimension of the a-th dimension. The smaller the value, the greater the probability that the vth abnormal data in the a-th dimension is caused by a fault.
[0074] Furthermore, a preset probability threshold T is defined. ε When ε a,v ≥T ε When the time is specified, it indicates that the v-th abnormal data in the a-th dimension of the target system is fault data, that is, abnormal data generated by robot failure, meaning that the robot malfunctioned at the time corresponding to this fault data. The preset probability threshold T in this embodiment... ε =0.4, which is used as an example for description. Other values can be set in other implementations.
[0075] This concludes the embodiment.
[0076] The present invention also provides an underwater robot fault identification system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program stored in the memory to implement the steps of the aforementioned underwater robot fault identification method.
[0077] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for fault identification in underwater robots, characterized in that, The method includes the following steps: The underwater robot uses sensors inside to acquire a sequence of operational data in each dimension of each system during operation; each piece of operational data corresponds to a numerical value, a time, and the number of times the operation is performed. Based on the number of times each running data is run, each running data sequence is divided into several data segments; based on the values within each data segment, each data segment is further divided into several new data segments; based on the position of each new data segment within each data segment, the position of each running data within each new data segment, and the number of times each running data is run, several data combinations are obtained, each data combination containing running data of two different dimensions; based on the time differences of running data of different dimensions in each data combination and the numerical differences of running data of the same dimension in different data combinations, the probability that the value of running data of each dimension is an exponential power of the values of running data of other dimensions is obtained, thereby obtaining the correlation dimension of each dimension, the correlation relationship between each dimension and the correlation dimension, and the exponential relationship between the value of running data of each dimension and the value of running data of the correlation dimension. Based on the value of each running data, abnormal data in each dimension is obtained; based on the time of each abnormal data in each dimension and the time of each abnormal data in the correlation dimension, the comparison data of each abnormal data in each dimension in each correlation dimension is obtained; based on the abnormal data and comparison data in each dimension, the running data to the left of the abnormal data in the running data sequence of each dimension and the running data to the left of the comparison data, the correlation between each dimension and the correlation dimension, and the power relationship between the value of the running data in each dimension and the value of the running data in the correlation dimension, fault data is obtained, and anomaly detection is completed. The method for obtaining the power probability is as follows: In the formula, Indicates the first in the target system The value of the operational data in the first dimension is the first... The numerical values of the operational data in each dimension Power-order probability. Indicates the first in the target system The value of the operational data in the first dimension is the first... The numerical values of the operational data in each dimension Power-order probability. Indicates the first In the data combination, the first The numerical values of the operational data in each dimension. Indicates the first In the data combination, the first The numerical values of the operational data in each dimension. Indicates the first In the data combination, the first The numerical values of the operational data in each dimension Power of 1 Indicates the first In the data combination, the first The numerical values of the operational data in each dimension Power of 1 Indicates the first In the data combination, the first The numerical values of the operational data in each dimension Power of 1 Indicates the first In the data combination, the first The numerical values of the operational data in each dimension Power of 1 Indicates the first In the data combination, the first The numerical values of the operational data in each dimension. Indicates the first In the data combination, the first The numerical values of the operational data in each dimension. This represents the mean of the absolute values of the differences in time between two running data points across all data combinations. Indicates the first of all data combinations The numerical values of the operational data in each dimension The power minus the first element in the previous data combination The numerical values of the operational data in each dimension The difference of the power and the first power among all data combinations The operational data of each dimension minus the data of the previous data combination. The variance of the ratio of the differences in the operational data of each dimension. Indicates the first of all data combinations The numerical values of the operational data in each dimension The power minus the first element in the previous data combination The numerical values of the operational data in each dimension The difference of the power and the first power among all data combinations The operational data of each dimension minus the data of the previous data combination. The variance of the ratio of the differences in the operational data of each dimension. Indicates an index. This represents the preset maximum value of the exponent. It is an exponential function with the natural constant as its base.
2. The underwater robot fault identification method according to claim 1, characterized in that, The specific steps for dividing each running data sequence into several data segments based on the number of times each running data is executed are as follows: Taking the underwater robot's propulsion system as the target system, within the target system... In the operational data sequence of each dimension, operational data with the same number of runs are grouped into one data segment, and the target system's first... The operational data for each dimension is divided into several data segments.
3. The underwater robot fault identification method according to claim 2, characterized in that, The specific steps involved in dividing each data segment into several new data segments based on the values within the data segment are as follows: Within the target system The mode number of the numerical values of the operational data within any data segment in the operational data sequence of each dimension is denoted as the th digit of the target system when the power is fixed. The normal values for this data segment within each dimension; Remove the first from the target system In the operational data sequence of the [number] dimension, the value within this data segment represents the [number]th [value] in the target system when the power is fixed. The normal values of the data segment within the specified dimension are used to obtain the operating data of the target system. A continuous data interval within this data segment in a multi-dimensional running data sequence; Within the target system The continuous data interval within each data segment of the operational data sequence in the target system is denoted as the i-th... New data segments in each data segment of each dimension; Within the target system The new data segment within each data segment of the running data sequence in each dimension is denoted as the i-th segment. A new data segment in a running data sequence; Each new data segment corresponds to a number of runs and a position.
4. The underwater robot fault identification method according to claim 3, characterized in that, The process of obtaining several data combinations based on the position of each new data segment within each data segment, the position of each running data within each new data segment, and the number of times each running data is executed includes the following specific steps: In the The running data sequence and the first In a running data sequence, new data segments with the same number of runs and the same position are considered to be corresponding data segments; No. The first data segment in a new data sequence of the running data sequence The first running data and the first data segment in the corresponding data segment Each set of operational data corresponds to the others. The first The running data within a new data segment in a running data sequence and its corresponding data form a data combination, resulting in several data combinations. Based on the data combination in the first... The data is sorted by the time of each dimension of the running data from smallest to largest.
5. The underwater robot fault identification method according to claim 4, characterized in that, The specific steps for obtaining the correlation dimension of each dimension, the correlation between each dimension and the correlation dimension, and the power relationship between the numerical values of the operational data of each dimension and the numerical values of the operational data of the correlation dimension are as follows: Get the maximum The value and the maximum The value; If the maximum The value is greater than the maximum The value will be the largest The value is denoted as the first in the target system. The dimension and the first Correlation of operational data in each dimension , will be the largest corresponding , recorded as ; If the maximum The value is less than the maximum The value will be the largest The value is denoted as the first in the target system. The dimension and the first Correlation of operational data in each dimension , will be the largest corresponding Recorded as ; Preset correlation threshold ,if The first in the target system The dimension and the first These dimensions are interrelated. Indicates the first in the target system The dimension and the first The correlation of each dimension Indicates the first in the target system The numerical values of the operational data in the first dimension and the first dimension The power relationship of the numerical values of the running data in each dimension.
6. The underwater robot fault identification method according to claim 2, characterized in that, The specific steps for obtaining abnormal data in each dimension based on the value of each running data are as follows: When the robot is running normally, the first... Normal range of operating data in each dimension The first in the preset target system The first dimension Normal range of operating data for each correlation dimension , Indicates the first in the preset target system The minimum value of the running data in each dimension. Indicates the first in the preset target system The maximum value of the running data in each dimension. Indicates the first in the preset target system The first dimension The minimum value of the running data for each correlation dimension Indicates the first in the preset target system The first dimension The maximum value of the running data for each correlation dimension; Within the target system The data sequence of each dimension is not in the interval The running data within is denoted as the first. Abnormal data within each dimension; [This will affect] the target system's [number]th [dimension]. The first dimension The data sequence of each correlation dimension is not in the interval The running data within is denoted as the first. The first dimension Abnormal data within each relevance dimension.
7. The underwater robot fault identification method according to claim 1, characterized in that, The specific steps for obtaining the comparison data for each anomalous data point in each dimension and in each correlation dimension based on the time of each anomalous data point in each dimension and the time of each anomalous data point in the correlation dimension are as follows: Will with the The time closest to the first abnormal data point The first dimension Outlier data within the relevance dimension is denoted as the i-th. The first abnormal data point in the... Comparison data for each correlation dimension.
8. The underwater robot fault identification method according to claim 1, characterized in that, The process of obtaining fault anomaly data based on the anomaly data and control data for each dimension, the left-hand running data and control data for each dimension in the running data sequence, the correlation between each dimension and the correlation dimension, and the power relationship between the values of the running data for each dimension and the values of the running data for the correlation dimension, includes the following specific steps: In the formula, Indicates the first Within the dimension, the _ The possibility that the abnormal data was caused by a fault. Indicates the first in the target system The number of relevance dimensions. Indicates the first in the target system The dimension and the first The correlation relationships across the various correlation dimensions Indicates the first Within the dimension, the _ The value of the abnormal data. Indicates the first in the target system The first dimension of the running data sequence The value of the running data to the left of the abnormal data. Indicates the first Within the dimension, the _ The first abnormal data point in the... The numerical values of the comparison data for each correlation dimension. The value of the power, Indicates the first in the target system The first dimension Within the operational data sequence of the correlation dimension, the first... The first dimension The first abnormal data point in the... The numerical values of the data running on the left side of the comparison data for each correlation dimension. The value of the power, Indicates the first in the target system The numerical values of the operational data in the first dimension and the first dimension The power relationship of the numerical values of the operational data in each correlation dimension. Represents the normalization function. It is an exponential function with the natural constant as its base; Preset probability threshold ,when At that time, the first in the target system Within the dimension, the _ The abnormal data is fault data.
9. An underwater robot fault identification system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the computer program is executed by the processor, it implements the steps of the underwater robot fault identification method as described in any one of claims 1-8.