Building machine working state recognition method based on multi-source sensing information

Abstract

The building machine working state recognition method based on multi-source sensing information of the application aims at the problem of lacking process flow monitoring means in the existing building machine construction process and low working state recognition accuracy, steps: real-time acquisition of acceleration time history signals on the main force components in the building machine running process, real-time reading of the distance between the template and the building machine frame body, and the pressure and stroke data of the jacking cylinder; extraction of the root mean square value, average value, peak-to-peak value, entropy value and offset characteristic value of the acceleration time history signal, the measured average value of the distance, pressure and stroke as characteristic values, derivation of the characteristic combination vector under six working states of the building machine, training of the classification model under six working states of the building machine through the machine learning method; acquisition of the current measured acceleration time history signal, distance, pressure, stroke and the corresponding characteristic combination vector of the building machine, real-time comparison and recognition of the six working states of the current building machine by comparing the classification model trained.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of building engineering construction, in particular to a building machine working state recognition method based on multi-source sensing information. BACKGROUND

[0002] The building machine is a high-altitude operation platform for concrete structure widely used in high-rise and super high-rise building engineering construction, which can carry other construction equipment such as tower crane, distributing machine, construction hoist attachment device, and cooperatively climb construction following the vertical construction flow pace of the building structure main body, and has good overall construction efficiency. However, the structure and process of the multi-device integrated building machine are relatively complex, and the safety risk is large during the climbing process, so effective safety state monitoring measures of the building machine need to be taken.

[0003] At present, the safety state monitoring of the building machine mainly focuses on the mechanical state and environmental state of the structural member, and the safety monitoring of the process flow is less. The building machine needs to go through multiple process flows such as climbing, binding reinforcement, opening and closing mold, and pouring concrete every time the concrete structure main body is constructed one layer. The monitoring parts and components in different process flows have different emphases. It is impossible to effectively obtain the real-time working behavior of the building machine by relying on the traditional modeling method and information model updating means, and the accuracy of the working state recognition is not high due to many construction interference factors. SUMMARY

[0004] In view of the problems that the monitoring means of the process flow is lacking and the accuracy of the working state recognition is low during the construction process of the building machine of the existing high-rise and super high-rise building, the purpose of the present application is to provide a building machine working state recognition method based on multi-source sensing information.

[0005] The technical scheme adopted by the present application to solve the technical problem is as follows:

[0006] S1: the working state of the building machine is divided into six working states of stopping work, binding reinforcement, closing mold, opening mold, lifting and descending, and pumping, which are respectively represented as S A , S BB , S BH , S BT , S C , and S D , an acceleration sensor is arranged at the center position of the main horizontal force layer of the building machine, the acceleration sensor is fixedly arranged on the main force member of the building machine, at least one distance sensor is arranged between the formwork and the building machine frame body, and a pressure stroke sensor is arranged in the jacking cylinder of the building machine;

[0007] S2: In the six working states of the building machine, real-time data information of each sensor is read and a real-time data curve is established, the real-time data curve is divided into n data segments according to a time period Δt, and each sensor obtains a sensing signal of n data segments;

[0008] S3: Time domain, frequency domain and time-frequency domain characteristics of each data segment are analyzed, a root mean square value R rms , an average value , a peak-to-peak value R pp , an entropy value R H , and an offset characteristic value R cr are extracted as characteristic values of the acceleration time history signal R measured by the acceleration sensor in the time period Δt, and the measured average value of the distance L between the template and the building machine frame, the measured pressure P and the stroke H of the pressure stroke sensor in the time period Δt are used as characteristic values of the distance sensor;

[0009] S4: The characteristic vectors corresponding to the acceleration sensor, the distance sensor and the pressure stroke sensor are obtained, and the characteristic combination vectors F of all sensors in the six working states are obtained according to the actual process flow of the building machine;

[0010] S5: The characteristic combination vectors F of all sensors in the six working states of the building machine are trained by using a classifier, and the trained classification model is deployed in the information monitoring system of the building machine;

[0011] S6: The building machine enters the real-time recognition stage, each sensor reads the sensing signal in real time, and the current measured signals R c , L c , P c , H c and the corresponding characteristic combination vectors F c are obtained according to steps S2-S4, which are compared with the classification model trained in step S5 in real time, and the current working state of the building machine is determined, and the classification results of the six working states of the current measured signal are output respectively;

[0012] Wherein, n is the total number of the real-time data curve segmented according to the time period Δt;

[0013] Δt is the sampling interval time of each sensor;

[0014] R is the acceleration time history signal measured by the acceleration sensor in the time period Δt;

[0015] R rms is the root mean square value of R;

[0016] is the average value of R; ​

[0017] R pp peak-to-peak value of R;

[0018] R H entropy value of R;

[0019] R cr offset characteristic value of R;

[0020] L is the distance between the template measured by the distance sensor and the building machine frame body within the time period Δt;

[0021] P is the pressure data measured by the pressure stroke sensor of the jacking oil cylinder;

[0022] H is the stroke data measured by the pressure stroke sensor of the jacking oil cylinder;

[0023] respectively, the measured average value of the distance, pressure and stroke of all distance sensors and pressure stroke sensors within the time period Δt;

[0024] F is the characteristic combination vector of all sensors in six working states;

[0025] R c is the current measured acceleration time history signal in the building machine running state recognition process;

[0026] F c is the characteristic combination vector of the acceleration time history signal R c in the building machine running state recognition process;

[0027] L c is the current measured stroke data of the jacking oil cylinder in the building machine running state recognition process;

[0028] P c is the current measured pressure data of the jacking oil cylinder in the building machine running state recognition process.

[0029] The method for identifying the working state of the building machine based on multi-source sensing information comprises the following steps: first, the building machine process is divided into six working states, namely, stopping, binding, closing, opening, lifting and pumping; an acceleration sensor is used to obtain the acceleration time history signal of the main force component in the running process of the building machine; a distance sensor arranged between the formwork and the building machine frame body is used to measure the distance between the formwork and the building machine frame body in real time; a pressure stroke sensor arranged in the jacking cylinder of the building machine is used to read the pressure and stroke data of the jacking cylinder in real time; the root mean square value, the average value, the peak-to-peak value, the entropy value and the offset characteristic value of the acceleration time history signal, the measured average value of the distance between the formwork and the building machine frame body, and the measured average value of the pressure and stroke of the jacking cylinder are extracted as characteristic values, and the characteristic combination vector of the six working states of the building machine is obtained; a classification model of the six working states of the building machine is trained by a machine learning method; then, the current measured acceleration time history signal, distance, pressure, stroke and corresponding characteristic combination vector of the building machine are obtained; finally, the current measured acceleration time history signal, distance, pressure, stroke and corresponding characteristic combination vector of the building machine are compared with the trained classification model in real time, so as to identify the six key working states of the building machine in real time and accurately, accurately reflect the dynamic change of the building machine on site, enhance the self-discriminating ability of the building machine, improve the intelligent level, and ensure the safety of the equipment operation.

[0030] Further, in step S3, the acceleration sensor has m readings in the time period Δt, the i-th reading is R i , i = 1, 2, 3…m, the root mean square value R rms , the average value , and the peak-to-peak value R pp of the acceleration time history signal R measured by the acceleration sensor in the time period Δt are calculated as follows:

[0031]

[0032]

[0033] R pp = max(R i )-min(R i )

[0034] Further, in step S3, the data of each acceleration time history signal R after the mean value is removed is subjected to Fourier transform to obtain the frequency domain data Y R , and the power spectral density S R is calculated:

[0035]

[0036] According to the power spectral density S R Calculate the probability density P of each frequency point i Then according to the probability density P i Calculate the entropy value R H The calculation formula is as follows:

[0037]

[0038]

[0039] Further, in step S4: the acceleration sensor, distance sensor, pressure stroke sensor respectively corresponding feature vector is expressed as:

[0040]

[0041]

[0042]

[0043]

[0044] According to the actual process flow of the building machine, the feature combination vectors F of all sensors in six working states are obtained respectively;

[0045] F = [F1, F2, F3, F4] T

[0046] Further, step S6 also includes step S7:

[0047] S7: Through the current measured data The classification result obtained in step S6 is corrected as a correction parameter, the distance value a between the mold plate and the building machine frame under the closed mold state is set as the distance judgment threshold, and the final output state is determined:

[0048] The classification result is S A When L >= 0, H = 0, P = 0, the output state S A , otherwise output state E.

[0049] The classification result is S BB When L > a, H = 0, P = 0, the output state S BB , otherwise output state E.

[0050] The classification result is S BH When L = a, H = 0, P = 0, the output state S BH , otherwise output state E.

[0051] The classification result is S BTWhen L < a, H = 0, P = 0, output state S BT , otherwise output state E.

[0052] The classification result is S C When L < a, H > 0, P > 0, output state S C , otherwise output state E.

[0053] The classification result is S D When L = a, H = 0, P = 0, output state S D , otherwise output state E.

[0054] Wherein, E is an abnormality recognition state.

[0055] Further, the step S3 further comprises: setting an offset characteristic value R cr , given an offset amount δ of an acceleration time history signal R, that is, For the acceleration time history signal R, whenever (R i - δ) × (R i-1 - δ) < 0, R cr increases once;

[0056] k = l c / L max

[0057] R cr is an offset characteristic value of the acceleration time history signal R;

[0058] δ is an offset amount of the acceleration time history signal R;

[0059] k is an offset factor of the set , k ∈ (0, 1)

[0060] l c is a horizontal distance between two jacking cylinders adjacent to the acceleration sensor;

[0061] L max is a horizontal distance between two jacking cylinders farthest apart in the tower crane.

[0062] Further, the classifier in the step S5 can select KNN, SVM machine learning algorithm for classification. BRIEF DESCRIPTION OF DRAWINGS

[0063] Figure 1 is a flow chart of the tower crane working state recognition method based on multi-source sensing information of the application. DETAILED DESCRIPTION

[0064] Combined Figure 1The method for identifying the working state of the building machine based on multi-source sensing information is described, and the specific steps are as follows:

[0065] S1: According to the construction process of the standard layer of the super high-rise building core tube structure, the working state of the building machine is divided into four kinds of stop, construction, lifting and pumping, and is represented by S A , S B , S C , S D respectively; wherein, the stop state S A is that the building machine is statically placed on the core tube structure, and there is no personnel and equipment activity on the building machine; the construction state S B is that the construction personnel normally perform the operations such as binding reinforcement and disassembling and assembling formwork in the building machine; the lifting state S C refers to the action state of the building machine rising or descending through its own mechanical power system; the pumping state S D refers to the state that the concrete distributing machine deployed on the building machine transports concrete to the floor through the pumping equipment; further, the construction state S B can be subdivided into three working states of binding reinforcement, assembling formwork and disassembling formwork, the binding reinforcement state S BB is that the construction personnel starts to bind the wall reinforcement after the building machine is climbed in place; the assembling formwork state S BH is that the formwork is fixed on both sides of the wall reinforcement after the wall reinforcement is bound; the disassembling formwork state S BT is that the formwork is removed after the wall concrete is poured and reaches the strength requirement, so the building machine in this embodiment has the above six working states.

[0066] An acceleration sensor is arranged at the center position of the main horizontal force layer of the building machine, the acceleration sensor is fixed closely on the main force component of the building machine, and the sampling frequency of the acceleration sensor is not less than 100Hz; at least one distance sensor is arranged between the formwork and the frame body of the building machine, each distance sensor can measure the distance between the formwork and the frame body of the building machine in real time; a pressure stroke sensor is arranged in the jacking oil cylinder of the building machine, and the pressure and stroke data of the jacking oil cylinder are read through PLC, and the sampling frequency of the distance sensor and the pressure stroke sensor is not less than 1Hz; the acquisition instrument sends all the data collected by each sensor to the monitoring information system of the building machine;

[0067] S2: In the S A , S BB , S BH , S BT , S C , S DIn the six working states, the data information of each sensor is read in real time and a real-time data curve is established, the real-time data curve is divided into n data segments according to a time period Δt, and each sensor obtains n data segments of the sensing signal;

[0068] S3: analyzing the time domain, frequency domain and time-frequency domain characteristics of each data segment, extracting the root mean square value R rms , the average value , the peak-to-peak value R pp , the entropy value R H , and the offset characteristic value R cr as the characteristic values; for the distance L between the template and the building machine frame body measured by the distance sensor, the pressure P and the stroke H measured by the pressure stroke sensor in the time period Δt, the measured average values in the time period Δt are used as the characteristic values;

[0069] S4: obtaining the characteristic vectors corresponding to the acceleration sensor, distance sensor and pressure stroke sensor respectively, and obtaining S A , S BB , S BH , S BT , S C , S D the feature combination vectors of all sensors in the six working states;

[0070] S5: training S A , S BB , S BH , S BT , S C , S D the feature combination vectors of all sensors in the six working states, and deploying the trained classification model in the information monitoring system of the building machine;

[0071] S6: the building machine enters the real-time identification stage, each sensor reads the sensing signal in real time, obtains the current measured signal R c , L c , P c , H c and the corresponding feature combination vector F c , compares it with the classification model trained in step S5 in real time, and judges the current working state of the building machine, and outputs S A , S BB , S BH , S BT , S C , S D ​Classification results of six working states

[0072] Wherein, n is the total number of the real-time data curve segmented according to the time period Δt;

[0073] Δt is the sampling interval time of each sensor, which is a constant, such as 10s, 20s, 30s, etc.

[0074] R is the acceleration time history signal measured by the acceleration sensor within the time period Δt;

[0075] R rms is the root mean square value of R;

[0076] is the average value of R;

[0077] R pp is the peak-to-peak value of R;

[0078] R H is the entropy value of R;

[0079] R cr is the offset characteristic value of R;

[0080] L is the distance between the template and the building frame body measured by the distance sensor within the time period Δt;

[0081] P is the pressure data measured by the pressure stroke sensor of the jacking oil cylinder;

[0082] H is the stroke data measured by the pressure stroke sensor of the jacking oil cylinder;

[0083] respectively, are the measured average values of the distance, pressure and stroke of all distance sensors and pressure stroke sensors within the time period Δt;

[0084] F is the feature combination vector of all sensors under six working states;

[0085] R c is the currently measured acceleration time history signal in the building machine running state recognition process;

[0086] F c is the feature combination vector of the acceleration time history signal R c in the building machine running state recognition process;

[0087] L c is the currently measured stroke data of the jacking oil cylinder in the building machine running state recognition process;

[0088] P c is the currently measured pressure data of the jacking oil cylinder in the building machine running state recognition process.

[0089] The building machine working state recognition method based on multi-source sensing information of the application firstly divides the building machine process into six working states of stopping, binding, combining, demolding, lifting and pumping, uses an acceleration sensor to obtain the acceleration time history signal on the main force component in the running process of the building machine, uses a distance sensor arranged between the formwork and the building machine frame to measure the distance between them in real time, and uses a pressure stroke sensor arranged in the jacking cylinder of the building machine to read the pressure and stroke data of the jacking cylinder in real time; the root mean square value, the average value, the peak-to-peak value, the entropy value and the offset characteristic value of the acceleration time history signal, the measured average value of the distance between the formwork and the building machine frame, and the measured average value of the pressure and stroke of the jacking cylinder are taken as characteristic values, and the characteristic combination vector of the six working states of the building machine is obtained, and the classification model of the six working states of the building machine is trained by the method of machine learning; then, the current measured acceleration time history signal, distance, pressure, stroke and corresponding characteristic combination vector of the building machine are obtained; finally, they are compared with the trained classification model in real time, and the six key working states of the current building machine are identified in real time and accurately, the on-site process dynamic change of the building machine is accurately reflected, the self-discriminating ability of the working state of the building machine is enhanced, the intelligent level is improved, the safety of the equipment operation is ensured, and the problems of lack of real-time monitoring means and low working state recognition accuracy of the building machine in the service process of high-rise and super high-rise buildings are solved.

[0090] In the above step S3, assume that the acceleration sensor has m readings in the time period Δt, the i-th reading is R i , i = 1, 2, 3…m, the root mean square value R rms , the average value and the peak-to-peak value R pp of the acceleration time history signal R measured by the acceleration sensor in the time period Δt are calculated as follows:

[0091]

[0092]

[0093] R pp = max(R i )-min(R i )

[0094] In the above step S3, the data of each acceleration time history signal R after removing the mean value is subjected to Fourier transform (FFT) to obtain the frequency domain data Y R , and the power spectral density S R is calculated:

[0095]

[0096] According to the power spectral density SR Calculate the probability density P of each frequency point i Then, according to the probability density P i Calculate the entropy value R H The calculation formula is as follows:

[0097]

[0098]

[0099] In the above step S4, the feature vectors corresponding to the acceleration sensor, distance sensor and pressure stroke sensor are respectively represented as:

[0100]

[0101]

[0102]

[0103]

[0104] According to the actual process flow of the building machine, the feature combination vectors F of all sensors in six working states are respectively obtained: A BB BH BT C D

[0105] F = [F1, F2, F3, F4] T

[0106] F1, F2, F3, F4 are the feature vectors corresponding to the acceleration sensor, distance sensor and pressure stroke sensor respectively;

[0107] F is the feature combination vector of all sensors in six working states.

[0108] In order to improve the reliability of the building machine working state recognition, the above step S6 further includes step S7, the specific content is as follows:

[0109] S7: According to the characteristics of the distance sensor and the pressure stroke sensor in the actual working state of the building machine, the The classification result is corrected by taking the distance value a between the mold plate and the building machine frame in the closed mold state as the distance judgment threshold value, and the final output state is determined:

[0110] The classification result is S A When L >= 0, H = 0, P = 0, the output state S A ​​​​​​Otherwise, output state E.

[0111] Classification result is S BB When L > a, H = 0, P = 0, output state S BB Otherwise, output state E.

[0112] Classification result is S BH When L = a, H = 0, P = 0, output state S BH Otherwise, output state E.

[0113] Classification result is S BT When L < a, H = 0, P = 0, output state S BT Otherwise, output state E.

[0114] Classification result is S C When L < a, H > 0, P > 0, output state S C Otherwise, output state E.

[0115] Classification result is S D When L = a, H = 0, P = 0, output state S D Otherwise, output state E.

[0116] Wherein, E is an abnormality recognition state, that is, a state that does not occur simultaneously in the actual construction process of the building machine.

[0117] The step S3 further comprises: due to the different step of the jacking oil cylinder, the measurement value of the acceleration sensor in the lifting state of the building machine will be offset, in order to improve the feature recognition rate of the lifting state of the building machine, and avoid misjudgment due to the interference of construction noise, an offset feature value R cr , which is the number of times of intersection of the acceleration time history signal waveform and the time axis, and a given offset amount δ of the acceleration time history signal R, that is For the acceleration time history signal R, every time (R i - δ) × (R i-1 - δ) < 0, R cr increases once;

[0118] k = l c / L max

[0119] R cr is the offset feature value of the acceleration time history signal R;

[0120] δ is the offset amount of the acceleration time history signal R;

[0121] k is the offset factor of R , k ∈ (0, 1)

[0122] lc L is the horizontal distance between two jacking cylinders adjacent to the acceleration sensor;

[0123] L max L is the horizontal distance between two jacking cylinders farthest apart in the building machine.

[0124] The classifier in the step S5 can select KNN, SVM, etc. machine learning algorithm for classification.

[0125] The above description is only a description of the preferred embodiments of the present application, and is not any limitation on the scope of the present application, any modification made by the person skilled in the art according to the above disclosure belongs to the scope of claims.

Claims

1. A method for identifying the working status of a building construction machine based on multi-source sensor information, characterized in that, The steps are as follows: S1: The working states of the building construction machine are divided into six states: stop, reinforcement binding, formwork closing, formwork removal, lifting, and pumping, denoted as S1. A S BB S BH S BT S C S D An acceleration sensor is installed at the center of the main horizontal load-bearing layer of the building machine. The acceleration sensor is fixed in close contact with the main load-bearing component of the building machine. At least one distance sensor is installed between the template and the frame of the building machine. A pressure stroke sensor is installed in the lifting cylinder of the building machine. S2: Under the six working states of the building machine, the data information of each sensor is read in real time and a real-time data curve is established. The real-time data curve is divided into n data segments according to the time period Δt. Each sensor obtains the sensing signal of n data segments respectively. S3: Analyze the time domain, frequency domain, and time-frequency domain characteristics of each data segment. For the acceleration time history signal R measured by the accelerometer within the time period Δt, extract the root mean square value R of that data segment. rms ,average value Peak-to-peak value R pp Entropy value R H , offset eigenvalue R cr As characteristic values, the distance L between the template and the building frame measured by the distance sensor, and the pressure P and stroke H measured by the pressure-stroke sensor within the time period Δt are all the measured average values ​​within the time period Δt. As an eigenvalue; Set the offset characteristic value R cr Given an acceleration time history signal R with an offset δ, i.e. For this acceleration time history signal R, whenever (R i -δ)×(R i-1 When -δ)<0, R cr Add one more time; k=l c / L max R cr The offset characteristic value of the acceleration time history signal R; δ is the offset of the acceleration time history signal R; k is a set value. The offset factor, k∈(0,1) l c The horizontal distance between the two lifting cylinders adjacent to the acceleration sensor; L max This refers to the horizontal distance between the two farthest lifting cylinders in a building construction machine. R i The instantaneous value of the acceleration time history signal R collected at the i-th sampling moment within the time period Δt; S4: Obtain the feature vectors corresponding to the acceleration sensor, distance sensor, and pressure stroke sensor respectively. Based on the actual process flow of the building construction machine, obtain the feature combination vector F of all sensors under the six working states respectively. S5: Use a classifier to train the feature combination vector F of all sensors under the six working states of the building machine, and deploy the trained classification model in the information monitoring system of the building machine. S6: The building machine enters the real-time identification stage. Each sensor reads the sensing signal in real time and obtains the current measured signal R according to steps S2 to S4. c L c P c H c and the corresponding feature combination vector F c The model is compared in real time with the classification model trained in step S5 to determine the current working state of the building machine, and the classification results of the six working states of the current measured signal are output respectively; among them, F c Acceleration time history signal R during the building construction machine operation status identification process c The feature combination vector.

2. The method for identifying the working status of a building construction machine based on multi-source sensor information according to claim 1, characterized in that, In step S3: Assume that the accelerometer has a total of m readings within the time period Δt, and the i-th reading is R. i Let i = 1, 2, 3…m, and Rm be the root mean square value of the acceleration time history signal R measured by the accelerometer within the time interval Δt. rms ,average value Peak-to-peak value R pp The calculation formula is as follows:

3. The method for identifying the working status of a building construction machine based on multi-source sensor information according to claim 1, characterized in that, In step S3: the mean-free data of each acceleration time history signal R is subjected to Fourier transform to obtain frequency domain data Y. R And calculate the power spectral density S R : According to the power spectral density S R Calculate the probability density P at each frequency point i Then, based on the probability density P i Calculate the entropy value R H The calculation formula is as follows:

4. The method for identifying the working status of a building construction machine based on multi-source sensor information according to claim 1, characterized in that, In step S4: the feature vectors corresponding to the acceleration sensor, distance sensor, and pressure stroke sensor are respectively represented as follows: Based on the actual process flow of the building construction machine, the feature combination vector F of all sensors under six working states is obtained respectively; F=[F1,F2,F3,F4] T 。 5. The method for identifying the working status of a building construction machine based on multi-source sensor information according to claim 1, characterized in that, Step S6 is followed by step S7: S7: Based on the current measured data The classification result obtained in step S6 is corrected as a correction parameter. The distance value 'a' between the template and the building frame in the mold-closed state is set as the distance judgment threshold to determine the final output state: The classification result is S A When L>=0, H=0, P=0, the output state S A Otherwise, output state E; The classification result is S BB When L>a, H=0, P=0, the output state S BB Otherwise, output state E; The classification result is S BH When L = a, H = 0, and P = 0, the output state S BH Otherwise, output state E; The classification result is S BT When L < a, H = 0, and P = 0, output the status S BT Otherwise, output the status E; The classification result is S C When L<a,H> When P > 0, output state S C Otherwise, output state E; The classification result is S D When L = a, H = 0, and P = 0, the output state S D Otherwise, output state E; Where E represents the anomaly detection status.

6. The method for identifying the working status of a building construction machine based on multi-source sensor information according to claim 1, characterized in that, In step S5, the classifier can be either KNN or SVM machine learning algorithms for classification.

Images