Online diagnosis method of health status of impact piston of rock drilling rig based on time series characteristics

Abstract

The present invention discloses an online diagnosis method for the health status of the impact piston of a rock drilling rig based on time series characteristics. The method comprises the following steps: firstly, the sensor signal associated with the impact motion of the impact piston is obtained by means of a pressure sensor and a speed sensor installed in the hydraulic system of the rock drilling rig, thereby constructing an original data set; then, feature extraction is performed on the original data set, a filtered feature data set is obtained and input into an LSTM neural network model, a future prediction time series statistical feature data set is obtained and input into a health status diagnosis fuzzy system, and an estimation of the health status of the impact piston is obtained, thereby determining the health status of the impact piston. The present invention uses easily accessible hydraulic system sensor signals, solves the current problem of online monitoring and diagnosis of the health status of the impact piston of a rock drilling rig through hydraulic system pressure and speed sensor data, reduces construction costs and problem diagnosis time, and improves the efficiency of impact piston status diagnosis.

Description

Technical Field

[0001] The present invention relates to an online diagnosis method for the health status of a drilling rig, which belongs to the field of intelligent operation and maintenance of key components of a drilling rig, and specifically relates to an online diagnosis method for the health status of an impact piston of a drilling rig based on a time series statistical feature fuzzy network. Background Art

[0002] At present, drilling and blasting is the main construction method for tunnel excavation projects in my country, and drilling rigs are one of the core construction equipment for drilling and blasting. Among the many components of a drilling rig, the impact piston is the core component of the drilling rig's impact operation. Once the impact piston fails, the efficiency of the entire drilling and blasting construction will be greatly affected, resulting in economic losses and even casualties.

[0003] However, on the one hand, in a complex and harsh construction environment, with various rock types and ever-changing geological conditions, the impact piston can easily suffer various mechanical damages such as end face wear, lead area damage, cavitation, etc., thus affecting its health status; on the other hand, the impact piston is located inside the rock drill, with a tight mechanism and a small space, making it difficult to directly observe its operating health status.

[0004] One of the commonly used maintenance methods for rock drilling rig impact pistons in industry is regular manual maintenance. This maintenance method is simple and direct, but has low efficiency, high labor costs, and may not be able to detect component health problems in a timely manner. The other is status diagnosis based on real-time monitoring of operation and maintenance parameters. In this method field, traditional model-based methods include the use of Kalman filtering algorithm, wavelet packet transform, improved Kalman filtering algorithm, etc., but it is difficult to establish accurate models in the field of complex and large equipment. With the rise of artificial intelligence, there are also many methods that try to use deep learning methods for neural network-based status diagnosis, but they either use too single data sources to fully reflect the status, or have too high requirements for sensor data, making it difficult to obtain data in actual working environments. Summary of the invention

[0005] In order to solve the above problems, the present invention provides an online diagnosis method for the health status of a rock drilling rig impact piston by integrating time series feature extraction and a fuzzy neural system.

[0006] The present invention uses the data signals that are easily collected during the operation of the drilling rig as the data basis, and constructs an online diagnosis method for diagnosing the health status of the impact piston of the drilling rig, including the fusion of time series statistical feature extraction and fuzzy neural system, so as to solve the problems currently arising in the process of diagnosing the health status of the impact piston of the drilling rig using hydraulic system sensor data monitoring.

[0007] The technical solution of the present invention is as follows:

[0008] Step 1: Obtain sensor signals associated with the impact movement of the impact piston through pressure sensors and velocity sensors installed in the hydraulic system of the drilling rig, thereby constructing the original data set;

[0009] Step 2: Preprocess the original data set to obtain the preprocessed feature data set;

[0010] Step 3: After inputting the preprocessed feature data to be predicted into the trained LSTM neural network model, the corresponding future prediction time series statistical feature data set is obtained;

[0011] Step 4: Input the future predicted time series statistical feature data set to be predicted into the health status diagnosis fuzzy system to obtain the health status estimate of the impact piston, determine the health status of the impact piston based on the health status estimate of the impact piston, and thus obtain the operating status of the impact piston of the drilling rig.

[0012] The step 2 is specifically as follows:

[0013] 2.1) After data cleaning, outlier removal and quartile-based data filtering of the original data set, a cleaned data set is obtained;

[0014] 2.2) Grouping the cleaned data set so that the data contained in each group is a drilling operation of a rock drilling machine of a rock drilling rig, thereby obtaining a grouped data set;

[0015] 2.3) Extract the time series statistical features from the grouped data set to obtain an initial feature data set, and then use Kalman filtering to filter the initial feature data set to obtain a preprocessed feature data set.

[0016] In the above 2.3), the time series statistical features are the maximum value, the average value and the root mean square value.

[0017] In step 4, the health status diagnosis fuzzy system includes a degradation degree data set U, a health status level vector V, a weight vector W and a membership matrix R. The weight vector W and the membership matrix R are multiplied to obtain a health status estimate.

[0018] The health status level vector V satisfies V=[v1, v2, v3, v4], where v1, v2, v3 and v4 represent the health status of the impact piston of the drilling rig as excellent, good, medium and poor, respectively.

[0019] The degradation degree data set U is obtained by subtracting the future real time series statistical feature data from the future predicted time series statistical feature data set;

[0020] The membership matrix R is calculated by substituting the degradation degree data set U into the membership functions of different health status levels.

[0021] Among the membership functions of different health status levels, the membership function formula for an excellent health status is as follows:

[0022]

[0023] The membership function formula for a good health status is as follows:

[0024]

[0025] The membership function formula for the health status is as follows:

[0026]

[0027] The membership function formula for poor health status is as follows:

[0028]

[0029] Among them, μ i1 (·),μ i2 (·),μ i3 (·) and μ i4 (·) represent the membership function values ​​when the health status is excellent, good, moderate and poor, respectively, and x i represents the i-th degradation degree data point, a ij is the jth fuzzy boundary of the i-th degradation degree data point, j = 1, 2, ..., 6.

[0030] In step 5, the health state corresponding to the maximum value in the health state estimation of the impact piston is used as the health state of the impact piston.

[0031] The beneficial effects of the present invention are:

[0032] The present invention uses easily accessible hydraulic system sensor signals, which solves the current problem of online monitoring and diagnosis of the health status of the impact piston of the rock drilling rig through hydraulic system pressure and speed sensor data, reduces construction costs and problem diagnosis time, and improves the efficiency of impact piston status diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings:

[0034] Figure 1 is the membership function of the improved fuzzy system used in the present invention;

[0035] Figure 2 It is a flow chart of the online diagnosis method of the health status of the impact piston of the rock drilling vehicle based on the fusion of time series feature extraction and fuzzy neural system used in the present invention;

[0036] Figure 3 Schematic diagram of the LSTM network structure used in the present invention;

[0037] Figure 4 Schematic diagram of the LSTM network unit used in the present invention;

[0038] Figure 5 It is the final diagnosis result of the embodiment used in the present invention. DETAILED DESCRIPTION

[0039] It should be noted that the embodiments of the present invention, that is, the features in the embodiments, can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0040] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0041] like Figure 2 As shown, the present invention comprises the following steps:

[0042] Step 1: The sensor signals associated with the impact movement of the impact piston are obtained by installing pressure sensors and speed sensors in the hydraulic system of the three-arm drilling rig, thereby constructing the original data set; the pressure sensor collects the impact pressure, and the speed sensor collects the rotation speed and propulsion speed.

[0043] Step 2: Preprocess the original data set to obtain the preprocessed feature data set;

[0044] Step 2 is as follows:

[0045] 2.1) After data cleaning, outlier removal and quartile-based data filtering of the original data set, a cleaned data set is obtained;

[0046] 2.2) Grouping the cleaned data sets according to the process of the drilling operation, so that the data contained in each group is a drilling operation of the rock drilling machine of the rock drilling rig, and obtaining the grouped data sets;

[0047] 2.3) Extract the time series statistical features from the grouped data set to obtain an initial feature data set, and then use Kalman filtering to filter the initial feature data set to obtain a preprocessed feature data set.

[0048] 2.3), the time series statistical features are the maximum value, average value and root mean square value.

[0049] There are many methods for statistical analysis of time series features, such as the impact pressure x of the i-th group of k data ij , count and calculate its maximum value, average value and root mean square value, where:

[0050] x imax =max(x ij )

[0051]

[0052]

[0053] In the formula, x imax 、x imean and x ivar are the maximum value, average value and RMS value of the striking pressure of the ith group, respectively; k is the total amount of data of the ith group; max(·) represents the maximum value function.

[0054] Step 3: Use the filtered feature data set to train the LSTM neural network model to obtain a trained LSTM neural network model;

[0055] Step 4: Construct a health status diagnosis fuzzy system;

[0056] In step 4, the health status diagnosis fuzzy system includes the degradation degree data set U, the health status level vector V, the weight vector W and the membership matrix R. The health status estimation is obtained by multiplying the weight vector W and the membership matrix R, where:

[0057] The health status level vector V satisfies V = [v1, v2, v3, v4], where v1, v2, v3 and v4 represent the health status of the impact piston of the drilling rig as excellent, good, medium and poor respectively;

[0058] In a specific implementation, the weight vector W is determined through experiments.

[0059] The degradation degree data set U is obtained by subtracting the future real time series statistical feature data from the future predicted time series statistical feature data set;

[0060] The membership matrix R is calculated by substituting the degradation degree data set U into the membership functions of different health status levels.

[0061] like Figure 1 As shown in the figure, among the membership functions of different health status levels, the membership function formula of the health status being excellent (v1) is as follows:

[0062]

[0063] The membership function formula for the health status of good (v2) is as follows:

[0064]

[0065] The membership function formula for the health status of medium (v3) is as follows:

[0066]

[0067] The membership function formula for the poor health status (v4) is as follows:

[0068]

[0069] Among them, μ i1 (·),μ i2 (·),μ i3 (·) and μ i4 (·) represents the membership function when the health status is excellent, good, moderate and poor, respectively, x i represents the i-th degradation degree data point, a ij is the jth fuzzy boundary of the i-th degradation degree data point, j = 1, 2, ..., 6.

[0070] The formula of the membership matrix R is as follows:

[0071]

[0072] The formula for estimating the health status F is: F = W × R, satisfying F = [f1, f2, f3, f4], where f1, f2, f3, f4 are estimated values ​​when the health status of the impact piston of the drilling rig is excellent, good, medium, and poor, respectively.

[0073] Step 5: After inputting the preprocessed feature data to be predicted into the trained LSTM neural network model, the corresponding future prediction time series statistical feature data set is obtained;

[0074] Step 6: Input the future predicted time series statistical feature data set to be predicted into the health status diagnosis fuzzy system to obtain the health status estimate of the impact piston, determine the health status of the impact piston based on the health status estimate of the impact piston, and thus obtain the operating status of the impact piston of the drilling rig.

[0075] In step 6, according to the health status level vector V, the health status corresponding to the maximum value in the health status estimation of the impact piston is used as the health status of the impact piston.

[0076] By using the easily accessible hydraulic system sensor signals through the above steps, the current problem of online monitoring and diagnosis of the health status of the impact piston of the drilling rig is solved, the construction cost and problem diagnosis time are reduced, and the efficiency of the impact piston status diagnosis is improved.

[0077] The following is an explanation in conjunction with an optional embodiment. In this embodiment, the following defects of the prior art are taken into account: (1) The existing method for diagnosing the health status of the impact piston of a rock drilling rig is mainly regular maintenance, that is, regularly disassembling the rock drill for direct inspection and maintenance. This maintenance method is simple and direct, but it is inefficient, has high labor costs, and may not be able to detect component health problems in a timely manner; (2) It is difficult to establish an accurate mathematical model for real-time monitoring of operation and maintenance parameters based on models in the field of complex large-scale equipment; (3) The existing artificial intelligence-based methods or sensor data are single and difficult to fully reflect the health status of components, or the sensors used are difficult to install on rock drilling rigs with harsh working environments; (4) The original hydraulic system sensor data has large noise and hidden features, making it difficult to directly use it to diagnose the health status of the impact piston of a rock drilling rig.

[0078] This embodiment proposes an online diagnosis method for the health status of the impact piston of a drilling rig, which includes time series statistical feature extraction and a fuzzy neural system, to solve the following problems: (1) It can diagnose the health status of the impact piston of a drilling rig online in real time; (2) Combined with a neural network, it can better fit the state of the impact piston of the drilling rig; (3) Use the pressure and speed sensor data of the hydraulic system to avoid the problem of difficult installation of traditional sensors; (4) Use a time series statistical method to mine the hidden features of the sensor data.

[0079] This embodiment proposes an online diagnosis method for the health status of the impact piston of a drilling rig, which includes time series statistical feature extraction and a fuzzy neural system. First, it is necessary to collect the pressure and speed sensor signals of the hydraulic system. Since the working environment of the impact piston of the drilling rig is harsh and it is located inside the rock drill of the drilling rig, the structure is complex and the space is small, it is impossible to directly install sensors to measure the vibration signal generated by the impact motion of the impact piston. Therefore, this example proposes to diagnose the health status of the impact piston of the drilling rig by indirectly measuring the hydraulic system sensor data related to the impact motion of the impact piston that is easier to measure. According to the motion analysis of the impact piston of the drilling rig, depth, feed speed, impact pressure, thrust pressure and rotation pressure are selected as monitoring data.

[0080] In the actual operation of the drilling rig, the operation time is relatively discontinuous, and the operation status of the drilling rig at each drilling distance is more concerned. Based on this consideration, this embodiment selects depth as the measurement base, starting from the stable operation of the impact piston of the drilling rig, records the current depth at every unit distance, and records the corresponding feed speed, impact pressure, thrust pressure and rotation pressure to form the original hydraulic system pressure and speed sensor data.

[0081] 1. Data preprocessing

[0082] Due to the harsh working environment of the drilling rig, which is usually accompanied by a lot of noise, there are often abnormal data records during the operation of the sensor. In this case, the original data has large noise, high complexity, and abnormal values, which is not conducive to subsequent processing. Therefore, the data is first preprocessed, including data cleaning, data feature extraction and data filtering, and finally a data set is formed.

[0083] Among them, data cleaning mainly includes the elimination of zero values, empty values ​​and the processing of abnormal values. Its purpose is to screen out the values ​​that do not contain the state characteristics of the impact piston of the drilling rig due to various reasons (loose installation, signal fluctuations, etc.).

[0084] After removing null values ​​and zero values, first divide the sliding window according to the operating cycle of the rock drilling rig, and use the quartile method to clean the data in each sliding window to remove abnormal values. The core of the quartile method is the quartile point, which in statistics refers to the arrangement of all values ​​from small to large and divided into four equal parts. The values ​​at the three dividing points are called the lower quartile (Q1 for short), and the values ​​at the 75% position are called the upper quartile (Q3 for short). When using the quartile method, the minimum and maximum estimated values ​​are constructed based on Q1 and Q3, and the calculation method is as follows:

[0085] x min =Q1-k(Q3-Q1)

[0086] x max =Q3+k(Q3-Q1)

[0087] Among them, x min represents the minimum estimate, x max represents the maximum estimate, k is a constant coefficient, and Q1 and Q3 represent the lower quartile and upper quartile mentioned above.

[0088] When using the quartile method to eliminate outliers, the maximum and minimum estimated values ​​are used as upper and lower limits. When the data at a certain point exceeds the limit, it is regarded as an outlier and replaced by the median in the sliding window. The calculation method of the median in the sliding window is as follows:

[0089] The data X1, X2, X3, X4...X in the sliding window N Arrange them in ascending order as X (1) , X (2) , X (3) , X (4) ...X (N) , N represents the number of data in the sliding window, then the median value X in the sliding window m for:

[0090]

[0091] 2. Data Grouping

[0092] After preprocessing the data, in order to further extract its time series statistical features, the data is grouped according to the drilling operation cycle of the drilling rig. Each cycle contains all the data of the drilling rig drilling a certain hole, and the grouped data is obtained. In this way, each group of data can contain the complete data generated by the drilling rig when operating in the hole. When the health status of the impact piston changes, the related sensor signals will change and be fully reflected in the grouped data.

[0093] 3. Data feature extraction

[0094] After grouping, the data is extracted for key features such as maximum value, average value, and root mean square. This can reduce the amount of data and time complexity on the one hand, and on the other hand, it can perform preliminary feature extraction and reduce the difficulty of subsequent time series prediction.

[0095] The extracted new data sequence is then Kalman filtered to further highlight its time series statistical characteristics. Kalman filtering is an algorithm that uses the linear system state equation to optimally estimate the system state through the system input and output observation data. Since the observation data includes the influence of noise and interference in the system, the optimal estimate can also be regarded as a filtering process. The optimal estimate consists of two parts, namely the prediction part and the update part. The former is that the Kalman filter predicts the data at the current moment through the data state at the previous moment, and the latter is that the Kalman filter adjusts the model prediction results according to the observation value at the current moment:

[0096] The first is the prediction process, in which the Kalman filter predicts the data at the current moment through the data state at the previous moment, which mainly includes two equations:

[0097]

[0098]

[0099] in, The state value at the previous moment The initial state prediction value generated, F is the state transfer matrix, B is the control matrix, u t-1 is the input for the previous moment; similarly, is the noise estimate P based on the previous moment t-1 The initial noise estimate generated at this moment, F is the corresponding state transfer matrix, and Q is the noise caused by the prediction model itself.

[0100] After the prediction process, we can get the predicted values ​​of the system state and noise distribution based on the state and noise distribution at the previous moment, but ignore the influence of the observation value at this moment. Therefore, we need to perform parameter update process in combination with the observation value at this moment to get the final prediction result. The update process is as follows:

[0101]

[0102]

[0103]

[0104] in, is the noise estimate P based on the previous moment t-1 The initial noise estimates generated at this moment, H and H T is the corresponding linear transformation matrix, R is the observation matrix, which is the covariance matrix composed of the covariance of the observation values, and K t The trade-off state matrix The parameter Kalman gain obtained with the observation matrix R is is the state value at the previous moment, z t is the mean of the observed value distribution, I is the unit matrix, is the noise estimate P based on the previous moment t-1 The initial noise estimate generated at this moment is based on K t Calculate the final state prediction result And the noise prediction result P t .

[0105] Combined with Kalman filtering, the maximum value, average value, and root mean square noise of the time series statistical features are filtered out, and finally the time series statistical feature data after Kalman filtering is obtained. This further removes the noise and highlights the time series statistical features. Subsequent methods all use the time series statistical feature data after Kalman filtering.

[0106] 4. LSTM prediction model construction

[0107] LSTM (Long Short-Term Memory Neural Network) is a neural network that is further improved on the traditional RNN (Recurrent Neural Network) and contains LSTM blocks. The traditional RNN records the current state of the model when the data is input in the hidden layer and assigns its weight at the output. Therefore, it is very good at processing time series data. However, when the data sequence is too long, the "memory" of the RNN network for the early time series data will be weakened. The LSTM network transforms the hidden layer space based on the RNN and adds gating units, so that it has better predictive capabilities for time series data. This embodiment uses the LSTM network model to train the extracted time series statistical feature data and form a key indicator regression model, such as Figure 3 As shown, the predicted values ​​of key indicators are output.

[0108] The LSTM network continuously updates itself by updating the cell states of the hidden layer. Compared with traditional recurrent neural networks such as RNN, LSTM has three channels, namely the forgetting channel, the input channel, and the output channel. During training, LSTM can use the σ function and the activation function tahn function to add or delete information to the cell state through these three channels, thereby updating the cell state.

[0109]

[0110]

[0111] The architecture of each unit of LSTM is as follows Figure 4 shown.

[0112] Among them, the forget channel determines whether the unit state information needs to be forgotten according to the following formula:

[0113] σ f =σ(W f [c t-1 ,x t ]+b f )

[0114] In the formula, x t is the current time input, c t-1 is the unit state at the previous moment, b f is the forgetting correction parameter. This formula is based on x t With c t-1 Output 0 or 1 to decide whether to forget or retain the previous state information h stored in the unit t-1 .

[0115] The principle of the input channel is similar to that of the forget channel, which is used to update the t-1 to h t The cell state of . Its equation is as follows:

[0116] σ i =σ(W i [c t-1 ,x t ]+b i )

[0117] In the formula, x t is the current time input, c t-1 is the unit state at the previous moment, b i To input correction parameters.

[0118] The final system state is determined by the forget and update channels:

[0119] h t =σ f ·h t-1 +σ i tanh(W c ·[c t-1 ,x t ]+b c )

[0120] In the formula, x t is the current time input, c t-1 is the unit state at the previous moment, b c It is the state correction parameter.

[0121] The output channel is used to determine the final output, and its equation is as follows:

[0122] σ0=σ(W o [c t-1 ,x t ]+b o )

[0123] In the formula, x t is the current time input, c t-1 is the unit state at the previous moment, b o Correction parameters for output.

[0124] In the construction of the LSTM prediction model, according to actual engineering experience, the health of the impact piston is most closely related to the impact pressure. Therefore, the time series statistical characteristic data of the impact pressure is selected as the key indicator, and the specific parameters are constructed as follows: the average value, maximum value and root mean square value of the key indicator impact pressure, a total of 3 dimensions, are used as output, and the average value, maximum value and root mean square value of the rotary pressure, thrust pressure, feed speed and interval time, a total of 12 dimensions, are used as input. Based on this input and output, the model hyperparameters are adjusted to complete the training and construction of the LSTM prediction model.

[0125] 5. Fuzzy system construction

[0126] The outputted time series statistical characteristics of the pressure relief pressure are highly correlated with the impact piston of the rock drilling rig. However, given the complex characteristics of the rock drilling rig system, it is difficult to directly use the time series statistical characteristics of the pressure relief pressure to qualitatively evaluate the health of the rock drilling rig impact piston. Therefore, the pressure relief pressure time series statistical characteristics predicted by LSTM are used as the benchmark value, and the pressure relief pressure time series statistical characteristics of the actual value and its residual are constructed to evaluate the degree of index degradation, and the health status of the rock drilling rig is evaluated in combination with the fuzzy system. In addition, according to the complex working conditions of the rock drilling rig impact piston, the traditional fuzzy system is improved to make it better fitting.

[0127] 6. Result Verification

[0128] The sensor data of 120 cycles measured online are processed to obtain the time series statistical characteristic data after Kalman filtering. The average value, maximum value and root mean square value of the rotary pressure, thrust pressure, feed speed and interval time, a total of 12 dimensions, are input into the model as input data to obtain the predicted values ​​of the average value, maximum value and root mean square value of the impact pressure, a total of 3 dimensions. The predicted values ​​are used as the baseline value and subtracted from the actual measured time series statistical characteristic data of the impact pressure to construct the residual, which is input into the fuzzy system to finally obtain its health status estimation, such as Figure 5 As shown. Among them, excellent means that the health status of the impact piston of the rock drilling rig is excellent, good means that the health status of the impact piston of the rock drilling rig is good, medium means that the health status of the impact piston of the rock drilling rig is medium, and poor means that the health status of the impact piston of the rock drilling rig is poor and needs to be treated. According to the trend of the curve in the figure, it can be judged that the impact piston of the rock drilling rig is in an excellent state most of the time, which is consistent with the actual situation, proving the feasibility and effectiveness of the method proposed in the present invention.

[0129] The above embodiments are used to illustrate the present invention rather than to limit the present invention. Any modification and change made to the present invention within the spirit of the present invention and the protection scope of the claims shall fall within the protection scope of the present invention.

Claims

1. An online diagnosis method for the health status of a rock drilling rig impact piston based on time series characteristics, characterized in that: The following steps are involved: Step 1: Obtain sensor signals associated with the impact movement of the impact piston through pressure sensors and velocity sensors installed in the hydraulic system of the drilling rig, thereby constructing the original data set; Step 2: Preprocess the original data set to obtain the preprocessed feature data set; Step 3: After inputting the preprocessed feature data to be predicted into the trained LSTM neural network model, the corresponding future prediction time series statistical feature data set is obtained; Step 4: Input the future prediction time series statistical feature data set to be predicted into the health status diagnosis fuzzy system to obtain the health status estimation of the impact piston, determine the health status of the impact piston based on the health status estimation of the impact piston, and thus obtain the operating status of the impact piston of the drilling rig; In step 4, the health status diagnosis fuzzy system includes a degradation degree data set U, a health status level vector V, a weight vector W and a membership matrix R. The health status estimation is obtained by multiplying the weight vector W and the membership matrix R. The membership matrix R is calculated by substituting the degradation degree data set U into the membership functions of different health status levels. The degradation degree data set U is obtained by subtracting the future real time series statistical feature data from the future predicted time series statistical feature data set.

2. The method for online diagnosis of the health status of the impact piston of a drilling rig based on time series characteristics according to claim 1 is characterized in that: The step 2 is specifically as follows: 2.1) After data cleaning, outlier removal and quartile-based data filtering of the original data set, a cleaned data set is obtained; 2.2) Grouping the cleaned data set so that the data contained in each group is a drilling operation of a rock drilling machine of a rock drilling rig, thereby obtaining a grouped data set; 2.3) Extract the time series statistical features from the grouped data set to obtain an initial feature data set, and then use Kalman filtering to filter the initial feature data set to obtain a preprocessed feature data set.

3. The method for online diagnosis of the health status of the impact piston of a drilling rig based on time series characteristics according to claim 2 is characterized in that: In the above 2.3), the time series statistical features are the maximum value, the average value and the root mean square value.

4. The method for online diagnosis of the health status of the impact piston of a drilling rig based on time series characteristics according to claim 1, characterized in that: The health status level vector V satisfies V=[v1, v2, v3, v4], where v1, v2, v3 and v4 represent the health status of the impact piston of the drilling rig as excellent, good, medium and poor, respectively.

5. The method for online diagnosis of the health status of the impact piston of a drilling rig based on time series characteristics according to claim 1, characterized in that: Among the membership functions of different health status levels, the membership function formula for an excellent health status is as follows: The membership function formula for a good health status is as follows: The membership function formula for the health status is as follows: The membership function formula for poor health status is as follows: Among them, μ i1 (·),μ i2 (·),μ i3 (·) and μ i4 (·) represent the membership function values ​​when the health status is excellent, good, moderate and poor, respectively, and x i represents the i-th degradation degree data point, a ij is the jth fuzzy boundary of the i-th degradation degree data point, j = 1, 2, ..., 6.

6. The method for online diagnosis of the health status of the impact piston of a drilling rig based on time series characteristics according to claim 1, characterized in that: In step 4, the health state corresponding to the maximum value in the health state estimation of the impact piston is used as the health state of the impact piston.

Images