Detection method of wheel pressure detection device for hoisting machinery based on multi-strain sensor fusion

Abstract

The present invention relates to a detection method of a wheel pressure detection device for a lifting machinery based on multi-strain sensor fusion, comprising a pattern and a strain gauge, wherein a clamp groove is provided on the upper surface of the pattern, and long grooves are provided on both sides of the bottom of the pattern, and a plurality of strain gauges are provided and arranged in a line in the long grooves, and slope surfaces are provided at both ends of the pattern, and arc ends are provided at the outer ends of the slope surfaces, and a clamp fixing is provided in the clamp groove. A wire management groove is provided between the long grooves on both sides, and the wire management groove is in a fishbone shape. The present invention can solve various problems caused by uneven wheel pressure of large wheels in actual detection equipment, and predict the approximate wheel pressure under the corresponding deformation of the pattern.

Description

Technical Field

[0001] The invention relates to the technical field of wheel pressure detection of lifting machinery, in particular to a detection method of a wheel pressure detection device of a lifting machinery based on multi-strain sensor fusion. Background Art

[0002] The wheel pressure of hoisting machinery is a key parameter in the design and manufacture of cranes. The maximum value of the wheel pressure of hoisting machinery not only affects the design of the overall structure of the hoisting machinery, but is also the main design basis for the track foundation. By detecting the wheel pressure of hoisting machinery, we can understand the specific situation of the load distribution of the equipment, make timely adjustments to the uneven load distribution, and avoid safety hazards such as overloading of hoisting machinery, asynchronous operation of trolleys, track climbing and excessive wear of wheels. The correct selection of equipment for hoisting machinery wheel pressure detection will help engineering practice, help further consolidate the crane evaluation business, and play a major role in crane track adjustment, whole machine safety, etc. The application market prospects are broad and can bring considerable results transformation business in the future. However, according to traditional devices, various problems caused by uneven wheel pressure of trolleys in actual engineering cannot be solved. Summary of the invention

[0003] The purpose of the present invention is to provide a detection method for a hoisting machinery wheel pressure detection device based on multi-strain sensor fusion, which can solve various problems caused by uneven wheel pressure of large wheels in actual detection equipment and predict the approximate wheel pressure under the corresponding deformation of the pattern.

[0004] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a wheel pressure detection device for lifting machinery based on multi-strain sensor fusion, comprising a pattern and a strain gauge, the upper surface of the pattern is provided with a clamp groove, and both sides of the bottom of the pattern are provided with long grooves, a plurality of strain gauges are arranged and arranged in a line in the long groove, the left and right ends of the pattern are provided with slope surfaces, the outer ends of the slope surfaces are provided with arc ends, a clamp fixing part is provided in the clamp groove, and a wire management groove is provided between the long grooves on both sides, and the wire management groove is in a fishbone shape.

[0005] Furthermore, the clamp fixing member includes a U-shaped clamp plate and a wedge block, a slot is provided in the vertical direction at the lower end of the U-shaped clamp plate, a first threaded hole is provided on the wedge block, a first bolt is sleeved in the first threaded hole, and the wedge block is locked by the first bolt and nut to fix the wedge block on the slot.

[0006] Furthermore, two second threaded holes are provided on both sides of the U-shaped fixture plate, and a top groove opposite to the second threaded holes is provided on the wedge block, and a second bolt locked with the second threaded holes is provided on the top groove.

[0007] A detection method for a hoisting machinery wheel pressure detection device based on multi-strain sensor fusion, the method comprising the following steps:

[0008] Step S1, adjusting the relative position between the pattern and the wheel pressure head to be located at the position where data is required to be collected, and obtaining data of each strain gauge at the position to be measured of each wheel relative to the pattern;

[0009] Step S2, extracting 80% of the data of each strain gauge measured at each position as a training set and 20% of the data as a test set, and then normalizing the data;

[0010] Step S3: After the data processing is completed, the BP neural network is constructed using the matlab software, the data of each strain gauge measured at each position is trained, the BP neural network parameters are set, the number of hidden layer nodes is determined, and the BP neural network is trained. When the minimum error of the prediction target of the training set reaches 0.001, the neural network training stops;

[0011] Step S4: input the data in the test set into the trained BP neural network to predict the wheel pressure value of the lifting machinery.

[0012] Furthermore, the number of hidden layer nodes in step S3 is determined using the following formula:

[0013] ;

[0014] In the formula : Number of input layer nodes; : Number of nodes in the output layer; : An adjustment parameter between 1 and 10.

[0015] Furthermore, the step S1 is further as follows: adjusting the relative position between the pattern and the wheel pressure head to the position where the data needs to be collected, adjusting the load and holding time of the data to be collected on the press control computer; applying the load, and when the load reaches the set value, the press holds the pressure for 20 seconds; during the holding period, observing the value of the strain gauge, and recording the data of each strain gauge after the value stabilizes; after the press is unloaded, observing whether the value of the strain gauge returns to zero, if not, repeat the above steps and re-measure until the measurement of each position to be measured on the pattern is completed.

[0016] Furthermore, the positions to be measured are positions where the strain gauges on each pattern are located and positions between adjacent strain gauges.

[0017] Furthermore, the "relative position between the adjustment pattern and the wheel pressure head" is further: adjusting the wheel pressure head to the center, right alignment, left offset load of 5mm and left offset load of 10mm with the specimen.

[0018] The beneficial effects of the present invention are as follows: based on the strain-wheel pressure mechanical model, a strain detection device integrating pattern and sensor is developed, and the wheel pressure is indirectly detected through strain. Based on the neural network model, the strain data of the corresponding pattern under the wheel pressure change is trained, and the relationship between the wheel pressure and the pattern strain is inferred, so that the prediction effect of the BP neural network model finally established is good. The BP neural network model in the present invention can better fit the relationship between the wheel pressure and the strain of each part of the sample. When the wheel pressure is detected on the track of the on-site lifting machinery later, it is only necessary to climb the wheel of the trolley from one side of the sample and stop it steadily. No precise positioning is required. It is only necessary to collect 10 sensor data and then import the 10 data into the neural network model to obtain 1 wheel pressure prediction value. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a structural schematic diagram of the pattern of the present invention;

[0020] Figure 2 for Figure 1 The enlarged view of point A in the middle;

[0021] Figure 3 It is a front view of the pattern of the present invention;

[0022] Figure 4 For the present invention Figure 3 Bottom view in

[0023] Figure 5 For the present invention Figure 3 Cross-section along the middle BB direction;

[0024] Figure 6 It is a structural schematic diagram of the U-shaped fixture plate of the present invention;

[0025] Figure 7 It is a structural schematic diagram of the wedge block of the present invention;

[0026] Figure 8 is a cross-sectional view of a wedge of the present invention;

[0027] Fig. 9 for Figure 8 The cross-sectional view in the AA direction;

[0028] Fig.10 A schematic diagram of a wheel parking position in the present invention;

[0029] Fig.11 It is a schematic diagram of the BP neural network algorithm flow;

[0030] Fig.12 This is a simple structural diagram of the neural network model;

[0031] Fig.13This is a schematic diagram of the BP neural network training results of the present invention;

[0032] Fig.14 This is a schematic diagram of the prediction error rate of the test set of the present invention;

[0033] Fig.15 Schematic diagram of test set error;

[0034] Fig.16 Schematic diagram of the test set prediction results of the present invention.

[0035] Among them: 1. style, 2. long groove, 3. slope surface, 4. fixture fixing part, 41. U-shaped fixture plate, 42. wedge block, 43. slot, 44. first threaded hole, 45. second threaded hole, 46. top groove, 47. first bolt, 48. second bolt, 49. nut, 5. arc end, 6. cable management groove, 7. fixture groove, 8. track. DETAILED DESCRIPTION

[0036] The present invention will be further described below in conjunction with the accompanying drawings.

[0037] See also Figures 1 to 9 The present invention provides an embodiment: a wheel pressure detection device for a lifting machinery based on multi-strain sensor fusion, comprising a pattern 1 and a strain gauge, wherein a clamp groove 7 is provided on the upper surface of the pattern 1, and long grooves 2 are provided on both sides of the bottom of the pattern 1, a plurality of strain gauges are arranged and arranged in a line in the long grooves 2, a slope surface 3 is provided at both ends of the pattern 1, and an arc end 5 is provided at the outer end of the slope surface 3, and a clamp fixing part 4 is provided in the clamp groove 7. The length of the long slot 2 of pattern 1 can be designed to be 450mm. At this time, seven strain gauges can be placed in the long slot 2 on one side. The installation method of the resistance strain gauge is generally pasting. The resistance strain gauge is placed in the long slot 2. Preferably, the strain gauge in the long slot 2 on one side is designed to be 5, and the long slots 2 on the left and right sides are a total of 10. Pattern 1 can be fixed on the track through the clamp fixing part 4 and the clamp slot 7. When the crane wheel rolls over pattern 1, the strain gauges at different positions in the long slot 2 will obtain strain data at each position; the strain gauge is a welded strain gauge sensor, and the model of the welded strain gauge sensor is DH1102. A wire shunting groove 6 is set between the long slots 2 on both sides, and the wire shunting groove 6 is fishbone-shaped. The two sides of the wheel are easy to interfere with the strain gauge components, causing the strain gauge to break and fall off, and the wire shunting groove 6 can sort out the strain gauge sensor cable and the front end of the hidden strain gauge sensor, and the wire shunting groove 6 can make the pattern 1 have good structural strength.

[0038] Please continue reading Figures 1 to 9As shown, in one embodiment of the present invention, the clamp fixing member 4 includes a U-shaped clamp plate 41 and a wedge block 42. A slot 43 is provided in the vertical direction at the lower end of the U-shaped clamp plate 41. A first threaded hole 44 is provided on the wedge block 42. A first bolt is sleeved in the first threaded hole 44. The wedge block 42 is locked by the first bolt and nut to fix the wedge block 42 on the slot 43. According to the cross-sectional shape of the track, the U-shaped clamp plate 41 is designed, and a slot 43 is provided on the pattern 1 to install the clamp fixing member 4. The pattern 1 is firmly fixed on the track through the clamping force of the clamp fixing member 4. The U-shaped clamp plate 41 is provided with a slot 43 in the vertical direction, so that the wedge block 42 is fixed to the slot 43 through the first bolt and nut. The wedge block 42 can move upward to clamp the track and fix the position of the pattern. The length of the opening slot of the pattern 1 is 60mm to 100mm, preferably 100mm.

[0039] Please continue reading Figures 1 to 9 As shown, in one embodiment of the present invention, two second threaded holes 45 are provided on both sides of the U-shaped fixture plate 41, and a top groove 46 is provided on the wedge block 42, which is opposite to the second threaded hole 45, and a second bolt locked with the second threaded hole is provided on the top groove. Two second threaded holes 45 with threads are drilled on both sides of the U-shaped fixture plate 41, and two top grooves 46 are milled on both sides of the wedge block 42. Two fixing bolts are added on both sides of the U-shaped fixture plate 41, and the bolts pass through the second threaded holes 45 and then support the groove of the wedge block 42, so that the wedge block 42 is clamped on the track, thereby clamping the pattern 1 on the track.

[0040] See also Figures 1 to 16 The present invention provides another embodiment: a method for detecting wheel pressure of a lifting machinery based on multi-strain sensor fusion, the method comprising the following steps:

[0041] Step S1, adjust the relative position between pattern 1 and the wheel pressure head to the position where data is required to be collected, and obtain the data of each strain gauge at the position to be measured of each wheel relative to pattern 1; when the number of strain gauges in the single-sided long groove 2 is designed to be 5, the wheel model has a total of 9 positions in the longitudinal direction relative to pattern 1, namely position 1, position 2, position 3, position 4, position 5, position 1-2 (between position 1 and position 2, and so on), position 2-3, position 3-4, position 4-5, as follows Fig.10 As shown;

[0042] Step S2, extracting 80% of the data of each strain gauge measured at each position as a training set and 20% of the data as a test set, and then normalizing the data;

[0043] Step S3: After the data processing is completed, the BP neural network is constructed using the matlab software, the data of each strain gauge measured at each position is trained, the BP neural network parameters are set, the number of hidden layer nodes is determined, and the BP neural network is trained. When the minimum error of the prediction target of the training set reaches 0.001, the neural network training stops;

[0044] Step S4: input the data in the test set into the trained BP neural network to predict the wheel pressure value of the lifting machinery.

[0045] Please continue reading Figures 1 to 16 As shown, in one embodiment of the present invention, the number of hidden layer nodes in step S3 is determined by the following formula:

[0046] ;

[0047] In the formula : Number of input layer nodes; : Number of nodes in the output layer; : An adjustment parameter between 1 and 10.

[0048] Please continue reading Figures 1 to 16 As shown, in one embodiment of the present invention, the step S1 is further as follows: adjusting the relative position between the pattern 1 and the wheel pressure head to be located at the position where data needs to be collected, and adjusting the load and holding time of the data to be collected on the press control computer; applying the load, and when the load reaches the set value, the press maintains the pressure for 20 seconds; during the holding period, observing the value of the strain gauge, and recording the data of each strain gauge after the value stabilizes; after the press is unloaded, observing whether the value of the strain gauge returns to zero, and if it does not return to zero, repeat the above steps and re-measure until the measurement of each position to be measured on the pattern 1 is completed.

[0049] Please continue reading Figures 1 to 16 As shown, in one embodiment of the present invention, the positions to be measured are the positions where the strain gauges on each pattern 1 are located and the positions between adjacent strain gauges.

[0050] Please continue reading Figures 1 to 16 As shown, in one embodiment of the present invention, the "relative position between adjustment pattern 1 and the wheel pressure head" is further: adjusting the position of the wheel pressure head and the sample to be centered, right-aligned, left-biased by 5mm and left-biased by 10mm. Specific embodiment 1:

[0051] Pressure testing machine name: microcomputer-controlled electronic universal testing machine; model: CMT5305; maximum force: 300 kN;

[0052] Stress and strain tester name: wireless strain sensor node; model: SG404EX; range: ±15000 µε;

[0053] The detailed steps of the experiment are as follows:

[0054] 1) Adjust the relative position between pattern 1 and the wheel pressure head to the position where data needs to be collected;

[0055] 2) Adjust the load and holding time of the data to be collected on the press control computer;

[0056] 3) Apply the load, and when the load reaches the set value, the press maintains the pressure for 20 seconds;

[0057] 4) During the pressure holding period, observe the strain gauge value, and record the data of each strain gauge after the value stabilizes;

[0058] 5) After the press is unloaded, observe whether the value of the strain gauge returns to zero. If it does not return to zero, repeat the above steps and re-measure.

[0059] The above is a complete measurement procedure for one load at one location, such as Fig.10 As shown in the figure, the upper position is left and the lower position is right. In the press test, the wheel model has a total of 9 test positions relative to the specimen in the longitudinal direction, namely position 1, position 2, position 3, position 4, position 5, position 1-2 (between position 1 and position 2, and so on), position 2-3, position 3-4, position 4-5. Position 1, position 2, position 3, position 4, and position 5 are directly above the 5 strain gauge patch positions. In order to simulate the diverse working conditions of the wheels of the on-site lifting machinery, the wheel model and the specimen have the conditions of centering, right alignment, left eccentric load 5mm and left eccentric load 10mm. The left eccentric load means that the right side of the wheel model is a certain distance away from the right edge of the specimen, that is, the specimen has a certain length that does not contact the wheel model. When the wheel is at these positions, 50kN, 100kN, 120kN, 150kN, 200kN, 250kN, and 290kN are applied respectively, and the microstrains measured by 10 strain gauges are recorded.

[0060] Test results:

[0061] A total of 250 sets of data were measured in the experiment. The recorded microstrain data are shown in Tables 1 to 4, where the header position represents the position of the wheel.

[0062] Table 1 Strain data results of wheel alignment experiment

[0063]

[0064] Table 2 Strain data results of wheel right alignment experiment

[0065]

[0066] Table 3 Experimental strain data results of wheel deflection 5mm to the left

[0067]

[0068] Table 4 Strain data results when the wheel deviates 10 mm to the left

[0069]

[0070] Construct BP neural network:

[0071] BP (Back Propagation) neural network is a multi-layer feedforward network trained by error back propagation. Its basic idea is gradient descent method, which uses gradient search technology to minimize the mean square error between the actual output value and the expected output value of the network. Its basic algorithm flow is as follows: Fig.11 shown.

[0072] The simple structure of BP neural network is as follows Fig.12 As shown, there is only one hidden layer.

[0073] The basic BP algorithm includes two processes: forward propagation of signals and backward propagation of errors. First, forward propagation is performed. After the data is input from the input layer, it is multiplied by the corresponding weights along the direction of the network, and then summed and biased. The result of the hidden layer is then used as input to calculate in the activation function, and the calculated result is subjected to the same operation to obtain the output result. The above completes one forward propagation.

[0074] After completing the forward propagation, the back propagation starts to update the parameters of each layer. The basic idea is to adjust the network parameters by calculating the error between the output layer and the expected value, so as to reduce the error. Then, the parameters in the network are updated according to the error function. The general methods for updating parameters include gradient descent method, stochastic gradient descent method, etc. Among them, the gradient descent method updates each parameter according to the descending direction of the gradient of the error function until the error is minimized.

[0075] Neural networks can discover patterns from large amounts of complex data with unknown patterns through continuous learning. The neural network method overcomes the complexity of the traditional analysis process and the difficulty of selecting the appropriate model function form. It is a natural nonlinear modeling process that does not require distinguishing what kind of nonlinear relationship exists, which greatly facilitates modeling and analysis. Today, neural network technology has demonstrated its extraordinary superiority in pattern recognition and classification, recognition filtering, automatic control, data prediction, etc.

[0076] Matlab software was used to construct the experimental data obtained by BP neural network training. A total of 250 sets of data were obtained in the experiment, of which 80% of the data were extracted as training sets and 20% of the data were used as test sets, that is, 200 sets of experimental data were used for neural network training, and 50 sets of data were used for prediction after training. After data classification is completed, the data needs to be normalized. The purpose of the processing is to limit the pre-processed data to a certain range, such as [0,1] or [-1,1], so as to eliminate the adverse effects caused by singular sample data, and normalization can speed up the speed of gradient descent to find the optimal solution, that is, to speed up the convergence of the training network.

[0077] After data preprocessing is completed, the BP neural network parameters are set. The number of hidden layer nodes can be determined by referring to the following empirical formula 1:

[0078] ;

[0079] In the formula : Number of input layer nodes; : Number of nodes in the output layer; : An adjustment parameter between 1 and 10.

[0080] According to the above empirical formula, the number of hidden layer nodes can be 5 to 13. According to multiple trainings, a better prediction result can be obtained by taking the number of hidden layer nodes as 10. After determining the number of hidden layer nodes, the neural network parameters are shown in Table 5.

[0081] Table 5 BP neural network training parameters

[0082] Training parameters Numeric Number of training sessions 5000 Learning Rate 0.01 Minimum training error 0.001 Number of hidden layers 3 The number of nodes in each hidden layer 10

[0083] The BP neural network training results are as follows Fig.13 As shown, the results show that in the 1921th round of training, the minimum error of the predicted target of the training set reached 0.001, and the neural network stopped training.

[0084] The prediction results of the test set are as follows: Fig.14 , Fig.15 , Fig.16 As shown. Fig.14 The test set prediction error rate chart shows that among the 50 test set data, most of the prediction error rates are Within, the prediction error rate of individual test data exceeds . Fig.13 The test set error graph shows that the prediction rate exceeds The data has an error of about 20 kN, and most of them are test data with a load of 100 kN or 120 kN. Since the strain values ​​of these two data are relatively close, the neural network has a poor prediction effect on the two. In the test set error graph, there are 7 data with an error of more than 10 kN. Fig.16 It can be seen from the test set prediction result graph that the true value curve and the predicted value curve are relatively close as a whole, with only a few data points having large deviations. It is not ruled out that the data is biased during the test process, resulting in poor prediction results of the neural network.

[0085] From the overall prediction effect of the neural network, it can be seen that the idea of ​​using the neural network to predict the wheel pressure value of the lifting machinery is feasible. The neural network can better fit the relationship between the wheel pressure and the strain of each part of the specimen, and the prediction effect is good. Although it can be observed in the prediction results of the test set that the prediction error rate of some data points is large, the algorithm can be further improved to optimize the prediction accuracy of the neural network.

[0086] Usually, finite element simulation analysis will be carried out after the test to compare and analyze the data of the two. However, in some cases, it is found that the test data and the simulation analysis data are quite different. For example, in laboratory tests, when the sample and the wheel are in the center position, the microstrain data on both sides are quite different, while in the simulation analysis under the same conditions, the microstrain data on both sides are close. This means that the setting conditions during the simulation analysis are different from the test conditions in the laboratory. The analysis found that because the bottom of the sample is an arc surface with a large radius, and the sample support of the laboratory press is a plane, the bottom of the sample was not fully supported during the laboratory test, and the contact form was line contact. During the simulation analysis, the support of the sample was the crane walking track, and the contact form of the bottom was surface contact. This resulted in the data of the two not being able to be compared. After the simulation analysis, the simulation data and the test data were relatively close when the sample and the wheel were in the centering condition.

[0087] The above description is only a preferred embodiment of the present invention and should not be construed as limiting the present application. All equivalent changes and modifications made within the scope of the patent application of the present invention should fall within the scope of the present invention.

Claims

1. A detection method for a hoisting machinery wheel pressure detection device based on multi-strain sensor fusion, characterized in that: The wheel pressure detection device of the lifting machinery includes a pattern and a strain gauge, the upper surface of the pattern is provided with a clamp groove, and the bottom two sides of the pattern are provided with long grooves, a plurality of strain gauges are provided, and are arranged in a line in the long groove, the left and right ends of the pattern are provided with slope surfaces, and the outer ends of the slope surfaces are provided with arc ends, and a clamp fixing piece is provided in the clamp groove, and a wire arranging groove is provided between the long grooves on both sides, and the wire arranging groove is in a fishbone shape; the clamp fixing piece includes a U-shaped clamp plate and a wedge block, and a slot is provided in the vertical direction of the lower end of the U-shaped clamp plate, and a first threaded hole is provided on the wedge block, and a first bolt is sleeved in the first threaded hole, and the wedge block is locked by the first bolt and a nut to fix the wedge block on the slot; The detection method of the hoisting machinery wheel pressure detection device comprises the following steps: Step S1, adjusting the relative position between the pattern and the wheel pressure head to be located at the position where data is required to be collected, and obtaining data of each strain gauge at the position to be measured of each wheel relative to the pattern; Step S2, extracting 80% of the data of each strain gauge measured at each position as a training set and 20% of the data as a test set, and then normalizing the data; Step S3: After the data processing is completed, the BP neural network is constructed using the matlab software, the data of each strain gauge measured at each position is trained, the BP neural network parameters are set, the number of hidden layer nodes is determined, and the BP neural network is trained. When the minimum error of the prediction target of the training set reaches 0.001, the neural network training stops; Step S4: inputting the data in the test set into the trained BP neural network to predict the wheel pressure value of the lifting machinery; The relative position between the adjustment sample and the wheel pressure head is further adjusted as follows: the wheel pressure head is adjusted to the center of the sample, right-aligned, left-biased 5mm and left-biased 10mm; The positions to be measured are the positions where the strain gauges on each pattern are located and the positions between adjacent strain gauges; The U-shaped fixture plate has two second threaded holes on both sides, the wedge block has a top groove opposite to the second threaded holes, and the top groove is provided with a second bolt locked with the second threaded holes; The number of hidden layer nodes in step S3 is determined using the following formula: ； In the formula : Number of input layer nodes; : Number of nodes in the output layer; : Adjustment parameter between 1-10; The step S1 further includes: adjusting the relative position between the pattern and the wheel pressure head to be located at the position where the data needs to be collected, and adjusting the load and holding time of the data to be collected on the press control computer; Apply the load, and when the load reaches the set value, the press maintains the pressure for 20 seconds; during the pressure maintenance period, observe the value of the strain gauge, and record the data of each strain gauge after the value stabilizes; after the press is unloaded, observe whether the value of the strain gauge returns to zero. If it does not return to zero, repeat the above steps and re-measure until the measurement of each position to be measured on the sample is completed.

Images