Roller Damage Classification and Identification Method and Device Based on Surface Waves and Neural Networks

Abstract

The present invention provides a method and device for classifying and identifying roll damage based on surface waves and neural networks. An ultrasonic measurement system is used to acoustically measure different types and degrees of defects or damages on the surface of a sample roll, and the measured signal features, corresponding defect or damage types and degrees obtained through analysis are used as training samples to construct a training database. Then, the data in the training database is used to train a BP neural network to obtain a roll surface damage classification and identification model capable of intelligently identifying defects and damages on the roll surface. Since ultrasonic surface waves are used to detect surface defect damages without damaging the roll to be measured, the method of the present invention is a non-destructive detection method. Since the training data covers data of various different types and degrees of defects or damages, the trained model can well identify various defects and damages on the surface of the roll to be measured and give corresponding degree information.

Description

Technical Field

[0001] The present invention belongs to the technical field of ultrasonic non-destructive testing, and particularly relates to a method and device for classifying and identifying roll damage based on surface waves and neural networks. Background Art

[0002] Rolls in the metallurgical industry are commonly used for rolling steel plates. They work under load for a long time, and cracks, indentations, and sticking of steel are likely to appear on the surface. At the same time, invisible damage or defects in the subsurface may also develop into open cracks. Defects or damages on the roll surface will batchwise affect the surface quality of the rolled steel plates, causing serious economic losses. Therefore, in the production and use process, effective non-destructive testing of the rolls, timely detection of defects or damages existing inside and on the surface, and additive repair, grinding, or replacement are essential technical means to improve product quality.

[0003] Non-destructive testing methods for roll defects or damages include direct observation method, visual recognition method, penetrant testing, magnetic particle testing, eddy current testing, and ultrasonic testing, etc. Direct observation is a common method for detecting roll defects or damages. This method mainly judges the degree and type of roll surface defects or damages based on the experience of engineers. However, reasons such as lack of experience of young engineers, human misjudgment or wrong judgment may all lead to incorrect test results. The above reasons are not conducive to the classification and identification of roll surface defects or damages. In addition, this method cannot detect the depth of defects or damages. The non-destructive testing method based on visual recognition mainly uses a camera instead of the human eye, saves manpower with machines, and takes pictures and identifies the defects or damages on the roll surface. However, in this technology, the viewing angle of the camera is small, restricted by the curved surface of the roll, and it is impossible to accurately identify the entire roll surface. Penetrant testing mainly utilizes the capillary phenomenon and the wetting effect of the penetrant. It is applicable to both metal and non-metal materials and has the advantages of simple operation and easy identification. Since it is difficult for the penetrant to enter defects or damages with relatively small sizes, penetrant testing is not sensitive to such damages. Magnetic particle testing mainly utilizes the bending of magnetic field lines at defects or damages to form a leakage magnetic field to adsorb magnetic particles, generating magnetic marks larger than the size of the defects or damages at discontinuous media. However, this method can only detect defects or damages on the surface or subsurface of ferromagnetic materials and is not sensitive to defects or damages inside the structure. Eddy current testing mainly utilizes the eddy current formed by electromagnetic induction, considers the influence of defects or damages on the eddy current, and obtains the degree and position of the defects or damages. But this method can only be used for conductive materials and is only sensitive to surface and subsurface defects or damages affected by the skin effect. Compared with the above methods, ultrasonic testing is applicable to both metal materials and non-metal materials and is not affected by the conductivity of the medium. Among them, the energy of ultrasonic surface waves is mainly concentrated on the surface of the object to be measured and is sensitive to surface and subsurface defects or damages with sizes approximately equal to the wavelength.

[0004] Existing ultrasonic testing methods are often used to determine whether there are damages or defects on the surface of the roll, but they cannot accurately identify the types of damages. For different types of damages, the repair processes (such as surfacing, induction hardening, thermal spraying technology, and laser surface modification, etc.) also vary greatly. Therefore, in order to improve the maintenance efficiency of the roll, a new method that can accurately identify the types and degrees of surface defects or damages of the roll is needed. Summary of the Invention

[0005] The present invention is made to solve the above problems, and the purpose is to provide a recognition method and corresponding device that can identify various defects or damages on the surface or subsurface of the roll without damage, automatically, and with high accuracy. The present invention adopts the following technical solutions:

[0006] The present invention provides a method for classifying and recognizing roll damages based on surface waves and neural networks, which is characterized by including the following steps:

[0007] Step S1: Use an ultrasonic measurement system to generate ultrasonic surface waves, perform acoustic measurement on the defects or damages on the surface of the sample roll, and obtain a measurement signal.

[0008] Step S2: Analyze the characteristics of the corresponding measurement signals for different types and degrees of defects or damages.

[0009] Step S3: Combine the characteristics of the measurement signals, the types and degrees of the corresponding defects or damages to form training samples, and store them in a training database.

[0010] Step S4: Establish a BP neural network model, and use the training database to train the BP neural network model to obtain the trained BP neural network model as a roll surface damage classification and recognition model.

[0011] Step S5: Use the ultrasonic measurement system to perform acoustic measurement on the roll surface, collect the measurement signal, and input the measurement signal into the roll surface damage classification and recognition model, so as to obtain the types and degrees of the defects or damages on the roll surface.

[0012] The roll damage classification and recognition device based on surface waves and neural networks provided by the present invention may further have the following technical features: The ultrasonic measurement system includes an arbitrary waveform generator, a power amplifier, a left ultrasonic transducer, a right ultrasonic transducer, a duplexer, an oscilloscope, a first filter, a second filter, a computer, and an ultrasonic transducer clamping device. The arbitrary waveform generator is connected to the power amplifier and the oscilloscope. The power amplifier is connected to the duplexer. The duplexer is connected to the first filter and the oscilloscope. The oscilloscope is further connected to the second filter and the computer. The left ultrasonic transducer is connected to the first filter. The right ultrasonic transducer is connected to the first filter and the second filter. Both the left ultrasonic transducer and the right ultrasonic transducer are mounted on the ultrasonic transducer clamping device.

[0013] The roll damage classification and recognition device based on surface waves and neural networks provided by the present invention may further have the following technical features: The ultrasonic transducer clamping device includes: a circumferential positioning module arranged on both sides of the roll, each having a turntable; an axial driving module arranged on the turntables of a pair of the circumferential positioning modules and capable of rotating along the outer periphery of the roll; and a pair of radial pressing modules movably arranged on the axial driving module. The left ultrasonic transducer and the right ultrasonic transducer are respectively mounted on a pair of the radial pressing modules and can be driven by the radial pressing modules to press towards the roll surface.

[0014] The roll damage classification and recognition method based on surface waves and neural networks provided by the present invention may further have the following technical features: Step S1 includes the following sub-steps:

[0015] Step S1-1: Install the sample roll in the ultrasonic transducer clamping device, adjust the positions of the left ultrasonic transducer and the right ultrasonic transducer so that the defect or damage is within their detection areas, excite ultrasonic surface waves with an initial excitation frequency through the left ultrasonic transducer and receive the reflected signals, and extract the first amplitude information of the reflected signals through the oscilloscope.

[0016] Step S1-2: Excite ultrasonic surface waves with an initial excitation frequency through the left ultrasonic transducer, receive the ultrasonic signals through the right ultrasonic transducer, and extract the second amplitude information of the ultrasonic signals through the oscilloscope.

[0017] Step S1-3: Excite ultrasonic surface waves through the left ultrasonic transducer and receive the reflected signals, and adjust the excitation frequency of the ultrasonic surface waves through the arbitrary waveform generator, gradually increasing from the first excitation frequency to the second excitation frequency at a predetermined step size, and extract the third amplitude information of the reflected signals through the oscilloscope during this process;

[0018] Step S1-4: Excite ultrasonic surface waves through the left ultrasonic transducer or the right ultrasonic transducer and receive the reflected signals, and extract the first arrival time information of the reflected signals through the oscilloscope;

[0019] Step S1-5: Adjust the position of the left ultrasonic transducer so that its distance from the right end of the roll is a predetermined distance. Excite ultrasonic surface waves through the left ultrasonic transducer and receive the reflected signals. Adjust the waveform height of the reflected signals in the oscilloscope to a predetermined height, and record the first gain of the power amplifier at this time;

[0020] Step S1-6: Adjust the position of the left ultrasonic transducer so that its distance from the defect or damage is the predetermined distance. Excite ultrasonic surface waves through the left ultrasonic transducer and receive the reflected signals. Adjust the waveform height of the reflected signals in the oscilloscope to a predetermined height, and record the second gain of the power amplifier at this time;

[0021] Step S1-7: Adjust the positions of the left ultrasonic transducer and the right ultrasonic transducer so that the defect or damage is within its detection area. Excite ultrasonic surface waves through the left ultrasonic transducer and receive the reflected signals, and extract the second arrival time information of the reflected signals through the oscilloscope;

[0022] Step S1-8: Excite ultrasonic surface waves through the left ultrasonic transducer, receive the reflected signals through the left ultrasonic transducer or the right ultrasonic transducer, and extract the third arrival time information of the reflected signals through the oscilloscope.

[0023] The roll damage classification and recognition method based on surface waves and neural networks provided by the present invention may further have the following technical features: the initial excitation frequency is 1 MHz, the first excitation frequency is 100 kHz, the second excitation frequency is 10.1 MHz, the predetermined step size is 2.5 MHz, and the step size can be adjusted according to actual situations. The predetermined height is 80% of the full screen height of the oscilloscope.

[0024] The roll damage classification and recognition method based on surface waves and neural networks provided by the present invention may further have the following technical features: the left end of the surface crack is point A, the tip is point B, and the right end is point C.

[0025] In step S1-4, the ultrasonic surface wave is first reflected at point A, and the reflected wave reaches the left ultrasonic transducer at time τ a and undergoes a second reflection and mode conversion at point B to generate an ultrasonic shear wave. The reflected wave of the second reflection passes through point A and reaches the left ultrasonic transducer at time τ b . The generated ultrasonic shear wave is reflected at the bottom surface of the specimen, converted into the ultrasonic surface wave at point B, passes through point A again, and reaches the left ultrasonic transducer at time τ c . The first time information includes τ a , τ b , and τ c . In step S1-7, the ultrasonic surface wave crosses the surface crack and reaches the right ultrasonic transducer at time τ d . At the same time, the ultrasonic surface wave passes through point A, point B, and point C and reaches the right ultrasonic transducer at time τ e . The ultrasonic surface wave passes through point A and undergoes a mode conversion at point B to generate an ultrasonic shear wave. The ultrasonic shear wave is reflected at the bottom surface of the specimen, passes through point B and point C, and reaches the right ultrasonic transducer at time τ f . The second time information includes τ d , τ e , and τ f .

[0026] The method for classifying and identifying roll damage based on surface waves and neural networks provided by the present invention may further have the following technical features. Among them, step S2 includes the following sub-steps:

[0027] Step S2-1: According to the first amplitude information, determine whether the maximum amplitude of the reflected signal in step S1-1 is greater than a predetermined value;

[0028] Step S2-2: If the determination in step S2-1 is no, further determine whether the decrease in the amplitude of the ultrasonic signal received by the right ultrasonic transducer in step S1-2 exceeds a predetermined threshold according to the first amplitude information and the second amplitude information;

[0029] Step S2-3: If the determination in step S2-2 is yes, the defect or damage is steel sticking;

[0030] Step S2-4: If the determination in step S2-1 is no, further determine whether the reflected signal disappears at some of the excitation frequencies in step S1-3;

[0031] Step S2-5: If the determination in step S2-4 is yes, the defect or damage is a buried defect or buried damage;

[0032] Step S2-6, if the judgment in Step S2-4 is negative and the defect is a surface crack.

[0033] The method for classifying and identifying roll damage based on surface waves and neural networks provided by the present invention may further have the following technical features. Among them, Step S2 further includes the following sub-steps:

[0034] Step S2-5a: Calculate the depth of the buried defect or the buried damage according to the excitation frequency when the reflection signal disappears.

[0035] Step S2-5b: Calculate the position information of the buried defect or the buried damage according to the first time information.

[0036] Step S2-6a: Judge whether the surface crack is an open crack.

[0037] Step S2-6b: If the judgment in Step S2-6a is negative, calculate the amplitude ratio of the reflection signal according to the second amplitude information, and obtain the depth of the surface crack according to the amplitude ratio.

[0038] Step S2-6c: If the judgment in Step S2-6a is positive, calculate the depth of the surface crack according to the second time information or the third time information.

[0039] Step S2-6d: Calculate the position information of the surface crack according to the first time information.

[0040] The method for classifying and identifying roll damage based on surface waves and neural networks provided by the present invention may further have the following technical features. Among them, in Step S2-5a, the depth of the buried defect or the buried damage is

[0041] In Step S2-5b, the distance between the buried defect or the buried damage and the left ultrasonic transducer is

[0042] In the formula, c R is the velocity of the ultrasonic surface wave in the sample roll, f x is the excitation frequency when the reflection signal disappears in Step S1-3, D is the distance between the left ultrasonic transducer and the right ultrasonic transducer.

[0043] In Step S2-6b, the difference Δ between the second gain and the first gain is:

[0044]

[0045]

[0046] In the formula, H Bis the height of the reflected wave of the right-angled side in the oscilloscope, H f is the height of the reflected wave of the surface crack in the oscilloscope, x is the distance between the left ultrasonic transducer and the surface crack or the right-angled side, L is the length of the surface crack, n is the ratio of the depth of the surface crack to the wavelength, a, b, c are related to the Poisson's ratio of the material of the roll, 2λ is the detection depth of the ultrasonic surface wave

[0047] In step S2-6c, the depth of the surface crack is or or is or

[0048] In the formula, c T is the velocity of the ultrasonic shear wave in the sample roll

[0049] In step S2-6d, the distance between the surface crack and the left ultrasonic transducer is

[0050] The present invention provides a roll damage classification and recognition device based on surface waves and neural networks, which is characterized by comprising: an ultrasonic measurement system for acoustically measuring the surface of the roll to obtain a measurement signal; and a roll surface damage classification and recognition model for identifying and classifying the measurement signal to obtain the corresponding defect or damage type and degree on the surface of the roll, wherein the roll surface damage classification and recognition model is a trained neural network model, and its training samples are obtained through the following steps: using the ultrasonic measurement system to acoustically measure the surface of a sample roll with defects or damages to obtain a measurement signal; analyzing the characteristics of the corresponding measurement signal for different types and degrees of defects or damages; and forming the training samples by combining the characteristics of the measurement signal, the corresponding defect or damage type and degree

[0051] Functions and effects of the invention

[0052] The present invention provides a method and device for classifying and identifying roll damage based on surface waves and neural networks. An ultrasonic measurement system is used to perform acoustic measurements on different types and degrees of defects or damages on the surface of a sample roll, and the measured signal features, corresponding defect or damage types and degrees obtained through analysis are used as training samples to construct a training database. Then, the data in the training database is used to train a BP neural network to obtain a roll surface damage classification and identification model capable of intelligently identifying defects and damages on the roll surface. Since ultrasonic surface waves are used to detect surface defect damages without damaging the roll to be measured, the method of the present invention is a non-destructive detection method; since the training data covers data of various different types and degrees of defects or damages, the trained model can well identify various defects and damages on the surface of the roll to be measured and give corresponding degree information. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 is a schematic diagram of the principle of the method for classifying and identifying roll damage based on surface waves and neural networks in an embodiment of the present invention;

[0054] Figure 2 is a flowchart of the method for classifying and identifying roll damage based on surface waves and neural networks in an embodiment of the present invention;

[0055] Figure 3 is a flowchart of step S1 in an embodiment of the present invention;

[0056] Figure 4 is a flowchart of step S2 in an embodiment of the present invention;

[0057] Figure 5 is a schematic structural diagram of the ultrasonic measurement system in an embodiment of the present invention;

[0058] Figure 6 is a usage state diagram of the ultrasonic transducer clamping device in an embodiment of the present invention;

[0059] Figure 7 is a schematic structural diagram of a BP neural network with a cognitive function in an embodiment of the present invention;

[0060] Figure 8 is a schematic distribution diagram of different types of defects or damages on the roll surface and subsurface in an embodiment of the present invention;

[0061] Figure 9 is a structural block diagram of the device for classifying and identifying roll damage based on surface waves and neural networks in an embodiment of the present invention.

[0062] Reference Signs:

[0063] Ultrasonic measurement system 10; Arbitrary waveform generator 11; Power amplifier 12; Duplexer 13; First filter 14a; Second filter 14b; Oscilloscope 15; Computer 16; Left ultrasonic transducer 17; Right ultrasonic transducer 18; Clamping device 20 for ultrasonic transducers; Circumferential positioning module 21; Support 211; First servo motor 212; Electromagnetic clutch 213; Planetary gear train 214; Disk 215; Support member 216; Axial drive module 22; Second servo motor 221; Guide rod 222; Lead screw 223; Radial pressing module 23; Pressing cylinder 231; Support platform 232; Roll 40; Roll damage classification and identification device 100 based on surface waves and neural networks; Ultrasonic measurement system 110; Intelligent classification and identification model 120 for roll surface damage; Control unit 130. Detailed implementation manner

[0064] In order to make the technical means, creative features, achieved purposes and effects of the present invention easy to understand, the following combines embodiments and drawings to specifically elaborate on the roll damage classification and identification method and device based on surface waves and neural networks of the present invention.

[0065] <Embodiment>

[0066] This embodiment provides a roll damage classification and identification method based on surface waves and neural networks, which is used to automatically identify defects or damages on the surface or subsurface of a roll, and the material of the roll surface layer is high-speed steel.

[0067] This method mainly includes the following steps: First, collect measurement signals of different types and degrees of defects or damages on the roll surface, and construct a training database; then establish a neural network model, and use the data in the training database to train the neural network model; finally, use the trained model to identify the types and degrees of roll defects or damages.

[0068] Figure 2 is a flowchart of the roll damage classification and identification method based on surface waves and neural networks in this embodiment.

[0069] As Figure 2 shown, the roll damage classification and identification method based on surface waves and neural networks specifically includes the following steps:

[0070] Step S1, use the ultrasonic measurement system to acoustically measure the defects or damages on the surface of the sample roll to obtain measurement signals.

[0071] Among them, the sample roll is a failed or discarded roll collected, or defects or damages are artificially prepared in the roll. Then, the types and degrees of the defects or damages in the sample roll are manually evaluated and recorded. Subsequently, for the defects or damages of different types and degrees in the sample roll, ultrasonic detection is carried out one by one using an ultrasonic measurement system.

[0072] When ultrasonic surface waves encounter steel sticking, indentation, surface cracks, and buried defects or damages on the surface or subsurface of the roll, corresponding physical phenomena such as reflection and attenuation will occur. Therefore, the types, positions, depths, etc. of the defects or damages can be judged based on the characteristics of the corresponding measurement signals.

[0073] Figure 5 It is a schematic structural diagram of the ultrasonic measurement system in this embodiment.

[0074] As Figure 5 shown, the ultrasonic measurement system 10 includes an arbitrary waveform generator 11, a power amplifier 12, a duplexer 13, a first filter 14a, a second filter 14b, an oscilloscope 15, a computer 16, a left ultrasonic transducer 17, and a right ultrasonic transducer 18.

[0075] The connection relationship is as Figure 5 shown. The arbitrary waveform generator 11 is connected to the power amplifier 12 and the oscilloscope 15. The power amplifier 12 is connected to the duplexer 13. The duplexer 13 is connected to the first filter 14a and the oscilloscope 15. The oscilloscope 15 is also connected to the second filter 14b and the computer 16.

[0076] The left ultrasonic transducer 17 is an ultrasonic transducer T / R and is connected to the first filter 14a by line ①. The right ultrasonic transducer 18 is an ultrasonic transducer T / R or an ultrasonic transducer R and is connected to the first filter 14a by line ② and to the second filter 14b by line ③.

[0077] First, an electrical signal is generated in an arbitrary waveform generator 11. The electrical signal then enters the left ultrasonic transducer 17 or the right ultrasonic transducer 18 successively through a power amplifier 12, a duplexer 13, and a filter 14a. Then, the left ultrasonic transducer 17 or the right ultrasonic transducer 18 is excited to convert the electrical signal into mechanical vibration, and the mechanical vibration will generate corresponding ultrasonic waves. Next, the ultrasonic waves pass through an acrylic wedge at a certain angle to form surface acoustic waves. When encountering defects or damages on the surface and subsurface of the roll, the surface acoustic waves will undergo corresponding physical phenomena such as reflection, attenuation, and mode conversion. The acoustic signal carrying defect or damage information passes through the acrylic wedge at the same angle and is converted into an electrical signal by the left ultrasonic transducer 17 and the right ultrasonic transducer 18. Finally, the electrical signal passes through the filter 14a and the duplexer 13 (lines ① and ②) or the filter 14b (line ③) and is displayed on an oscilloscope 15. At the same time, the computer 16 can also record relevant experimental data.

[0078] Among them, according to Snell's law, the angle of the acrylic can be obtained by formula (1):

[0079]

[0080] In the above formula, θ is the angle of the acrylic wedge, is the longitudinal wave velocity of the acrylic wedge, c R is the velocity of the surface acoustic wave in the roll.

[0081] The surface acoustic wave has non-dispersive properties, and the wave velocity is not affected by the frequency. When the Poisson's ratio of the material is less than 0.263, the ratio of the sound velocity of the surface acoustic wave to the shear wave is equal to:

[0082]

[0083] And when the Poisson's ratio of the material is greater than 0.263, the ratio of the propagation velocity of the surface acoustic wave to the shear wave is equal to:

[0084]

[0085] In the above formula:

[0086]

[0087] ζ = c T / c L (5)

[0088]

[0089]

[0090] In the above formula, ρ, E, and υ are the density, Young's modulus, and Poisson's ratio of the material, respectively. For example, in a roll with a high-speed steel outer layer material, its density, Young's modulus, and Poisson's ratio are 7700 kg / m 3 , 2.35×10 11 Pa and 0.27, respectively. Therefore, the propagation speeds of ultrasonic shear waves, longitudinal waves, and surface waves are 3466.34 m / s, 6175.46 m / s, and 3198.24 m / s, respectively. In this embodiment, the type of ultrasonic wave in the roll is judged according to the theoretical values of the propagation speeds of shear waves, longitudinal waves, and surface waves. The propagation speed during use is mainly based on the measured value.

[0091] Figure 6 is a schematic structural diagram of the ultrasonic transducer clamping device in this embodiment.

[0092] As Figure 6 shown, the ultrasonic measurement system 10 further includes a clamping device 20 for ultrasonic transducers (hereinafter referred to as the clamping device 20), which includes two sets of circumferential positioning modules 21, one set of axial driving modules 22, and two sets of radial pressing modules 23.

[0093] The circumferential positioning module 21 includes structures such as a bracket 211, a first servo motor 212, an electromagnetic clutch 213, a planetary gear train 214, a wheel disc 215, and a support member 216 for supporting the axial driving module 22. The planetary gear train 214 is rotatably installed on the bracket 211, the wheel disc 215 is connected to the planetary gear train 214, and the support member 216 is fixedly installed on the periphery of the wheel disc 215. The two sets of circumferential positioning modules 21 are respectively located on both sides of the roll 40 to be measured, and their wheel discs 215 are coaxial with the roll 40.

[0094] The axial driving module 22 includes structures such as a second servo motor 221, a guide rod 222, and a lead screw 223. The axial driving module 22 is respectively connected to the two sets of circumferential positioning modules 21 through the support components 216 on both sides, and the two ends of the guide rod 222 and the lead screw 223 are respectively installed on the two support members 216.

[0095] The radial pressing module 23 includes structures such as a pressing cylinder 231, a pressure sensor (not shown in the figure), and a support platform 232. The two support platforms are both installed on the guide rod 222 and the lead screw 223 and can move along the extension direction of the guide rod 222 and the lead screw 223. The two ultrasonic transducers (i.e., the left ultrasonic transducer 17 and the right ultrasonic transducer 18) are respectively installed at the end of the piston rod of the two pressing cylinders 231, and the output end of the ultrasonic transducer faces the roll 40. The pressure sensor is installed between the end of the piston rod of the pressing cylinder 231 and the ultrasonic transducer for detecting the pressure between the two. In addition, the pressing cylinder 231 is installed on the support platform 232 through a mounting bracket, and its mounting height is adjustable, that is, the height of the entire pressing structure is adjustable, so as to adapt to rolls 40 of various sizes.

[0096] Among them, the first servo motor 212 of the circumferential positioning module 21 on the left side of the figure is used to drive the planetary gear train 214, the planetary gear train 214 is used to drive the disk 215 to rotate, and the disk 215 is connected to the support member 216 of the axial drive module 22. Therefore, the axial drive module 22 and the radial pressing module 23 move circumferentially together with the disk 215; the electromagnetic clutch 213 of the circumferential positioning module 21 on the right side can limit the rotation of the planetary gear train 214, thereby hindering the circumferential movement of the axial drive module 22 and the radial pressing module 23, and ensuring the working angle of the ultrasonic transducer.

[0097] The second servo motor 221 of the axial drive module 22 is used to drive the lead screw 223 to drive the radial pressing module 23 at the left end to move axially. The guide rod 222 guides the two sets of radial pressing modules 23 at the same time, ensuring their movement directions.

[0098] The radial pressing module 23 presses the ultrasonic transducer against the roll 40. In this embodiment, the gas pressure of the piston rod of the pressing cylinder 231 is adjusted according to the value of the pressure sensor, so as to ensure that the pressing force of the ultrasonic transducer is constant.

[0099] According to the above structure, when the roll is installed in the clamping device 20, the left ultrasonic transducer 17 and the right ultrasonic transducer 18 are respectively located at its left and right ends, can move axially along the roll respectively, and can press against the roll.

[0100] Figure 3 It is the flowchart of step S1 in this embodiment.

[0101] As Figure 3 shown, based on the above ultrasonic measurement system 10, step S1 specifically includes the following sub-steps:

[0102] Step S1-1, install the sample roll in the clamping device 20, and adjust the positions of the left ultrasonic transducer 17 and the right ultrasonic transducer 18 so that the defects or damages on the roll surface are within their detection areas. Excite the ultrasonic surface wave with the initial excitation frequency through the left ultrasonic transducer 17 and receive the reflected signal, and extract the first amplitude information of the reflected signal through the oscilloscope 15.

[0103] The first amplitude information is used to judge the protrusion or depression of the surface defect or damage.

[0104] In this embodiment, in the clamping device 20, the first servo motor 212 that drives the circumferential positioning module 21 drives the planetary gear train 214 and the disk 215 to rotate, so that the axial drive module 22 rotates to a specific angle. At this time, the axial drive module 22, the radial pressing module 23, and the ultrasonic transducer are exactly above the defect or damage to be measured (above in the radial direction). Then, the second servo motor 221 that drives the axial drive module 22 adjusts the left ultrasonic transducer 17 and the right ultrasonic transducer 18 to both ends of the roll respectively. At this time, the defect or damage to be measured is within the detection range of both. Apply a coupling agent between the ultrasonic transducer and the roll, drive the pressing cylinder 231, press the ultrasonic transducer to the surface of the roll, and adjust the stroke of the piston rod of the pressing cylinder 231 according to the feedback of the pressure sensor to keep the value of the pressure sensor at about 75 N.

[0105] Then, using Line ①, the left ultrasonic transducer 17 is used to both excite and receive ultrasonic signals, and its initial excitation frequency is 1 MHz. The oscilloscope 15 and the computer 16 are used to extract the first amplitude information of the measured signal.

[0106] Repeat the above steps 5 times to obtain 5 pieces of first amplitude information.

[0107] In step S1-2, the left ultrasonic transducer 17 is used to excite ultrasonic surface waves at the initial excitation frequency, the right ultrasonic transducer 18 is used to receive ultrasonic signals, and the oscilloscope 15 is used to extract the second amplitude information of the ultrasonic signals.

[0108] The first amplitude information and the second amplitude information are used to judge the size of the convex defect or damage (such as steel sticking, etc.).

[0109] In this embodiment, the clamping device 20 is set as above, and then switch to Line ③. The left ultrasonic transducer 17 is used to excite ultrasonic signals, and its excitation frequency remains unchanged at 1 MHz. The right ultrasonic transducer 18 is used to receive ultrasonic signals. The oscilloscope 15 and the computer 16 are used to extract the second amplitude information of the measured signal.

[0110] Repeat the above steps 5 times to obtain 5 pieces of second amplitude information.

[0111] In step S1-3, the left ultrasonic transducer 17 is used to excite ultrasonic surface waves and receive reflected signals, and the arbitrary waveform generator 11 is used to adjust the excitation frequency of the ultrasonic surface waves, gradually increasing from the first excitation frequency to the second excitation frequency at a predetermined step size. And in this process, the oscilloscope 15 is used to extract the third amplitude information of the reflected signals.

[0112] The third amplitude information is used to judge whether there are buried defects or damages.

[0113] In this embodiment, the clamping device 20 is set as above, and then switch to line ①. The left ultrasonic transducer 17 is used to both excite and receive ultrasonic signals, and the oscilloscope 15 and the computer 16 are used to extract the third amplitude information of the measurement signal.

[0114] The excitation frequency of the left ultrasonic transducer 17 is adjusted to 100 kHz and gradually increased to 10.1 MHz in steps of 2.5 MHz, where the step size can be adjusted within the range of 0.5 MHz to 2.5 MHz according to the actual situation.

[0115] In this embodiment, the material of the roll surface layer is high-speed steel. At this time, the speed of the ultrasonic surface wave is 3198.2 m / s. When the excitation frequency is set to 100 kHz, the detection depth is 63.96 mm. When gradually adjusted to 10.1 MHz, the detection depth is 0.63 mm. Therefore, buried defects or buried damages within this depth range can be detected.

[0116] Repeat the above steps 5 times to obtain 25 pieces of third amplitude information.

[0117] In step S1-4, the ultrasonic surface wave is excited and the reflected signal is received through the left ultrasonic transducer 17 or the right ultrasonic transducer 18 in sequence, and the first arrival time information of the reflected signal is extracted through the oscilloscope 15.

[0118] The first arrival time information is used to calculate the specific position of the defect or damage.

[0119] In this embodiment, the clamping device 20 is set as above. Then first switch to line ①. The left ultrasonic transducer 17 is used to both excite and receive ultrasonic signals, and the oscilloscope 15 and the computer 16 are used to extract the arrival time τ of the reflected wave in the measurement signal g ; then switch to line ②. The right ultrasonic transducer 18 is used to both excite and receive ultrasonic signals, and the oscilloscope 15 and the computer 16 are used to extract the arrival time τ of the reflected wave in the measurement signal h . The first arrival time information includes τ g and τ h .

[0120] Repeat the above steps 5 times to obtain 10 pieces of time information.

[0121] In step S1-5, adjust the position of the left ultrasonic transducer 17 so that its distance from the right end of the roll is a predetermined distance. Excite the ultrasonic surface wave through the left ultrasonic transducer 17 and receive the reflected signal. Observe the waveform of the reflected signal in the oscilloscope 15, and adjust the height of the waveform to a predetermined height through the power amplifier 12, and record the first gain of the power amplifier 12 at this time.

[0122] In this embodiment, the axial driving module 22, the radial pressing module 23 thereon, and the left ultrasonic transducer 17 are rotated above the defect or damage to be measured. The second servo motor 221 is driven to move the left ultrasonic transducer 17 parallel to the surface of the rolling mill until the distance between the left ultrasonic transducer 17 and the right end of the rolling mill is equal to a predetermined distance. A coupling agent is applied between the ultrasonic transducer and the rolling mill. The stroke of the radial pressing module 23 on the left is adjusted to keep the value of the pressure sensor at about 75 N.

[0123] Then, switch to circuit ①. The left ultrasonic transducer 17 is used to both excite and receive ultrasonic signals simultaneously. Observe the waveform of the reflected wave on the oscilloscope 15. Adjust the height of the reflected wave to 80% max of the full screen of the oscilloscope 15 through the power amplifier 12, and record the first gain G3 of the power amplifier 12 at this time.

[0124] In step S1-6, adjust the position of the left ultrasonic transducer 17 so that its distance from the defect or damage is the predetermined distance. Excite the ultrasonic surface wave through the left ultrasonic transducer 17 and receive the reflected signal. Observe the waveform of the reflected signal on the oscilloscope 15, and adjust the height of the waveform to a predetermined height through the power amplifier 12, and record the second gain of the power amplifier 12 at this time.

[0125] For surface cracks with a depth less than 2λ, the first gain and the second gain are used to judge the actual depth of the surface crack.

[0126] In this embodiment, move the left ultrasonic transducer 17 to a distance equal to the predetermined distance from the defect or damage to be measured, and other settings are the same as above.

[0127] Then, use circuit ①. The left ultrasonic transducer 17 is used to both excite and receive ultrasonic signals simultaneously. Observe the waveform of the reflected wave on the oscilloscope 15. Adjust the height of the reflected wave to 80% max of the full screen of the oscilloscope 15 through the power amplifier 12, and record the second gain G4 of the power amplifier 12 at this time.

[0128] Repeat the above steps 5 times to obtain 10 pieces of gain information.

[0129] In step S1-7, adjust the positions of the left ultrasonic transducer 17 and the right ultrasonic transducer 18 so that the defect or damage is within their detection areas. Excite the ultrasonic surface wave through the left ultrasonic transducer 17 and receive the reflected signal, and extract the second time information when the reflected signal arrives through the oscilloscope 15.

[0130] For surface cracks with a depth greater than 2λ, that is, the single-probe method is used for measurement, and the measured second time information is used to calculate the depth of the open crack.

[0131] In this embodiment, the left ultrasonic transducer 17 is placed at the left end of the roll, and the defect or damage is within its measurement range. Other settings are the same as above. Then switch to circuit ①. The left ultrasonic transducer 17 is used to both excite and receive ultrasonic signals, and the oscilloscope 15 and the computer 16 are used to extract the arrival time τ of the reflected wave in the measurement signal. a , τ b and τ c .

[0132] Mark the left end, tip, and right end of the open crack as point A, point B, and point C respectively. The ultrasonic surface wave excited by the left ultrasonic transducer 17 undergoes the first reflection at point A, and the first reflected wave arrives at the left ultrasonic transducer 17 at time τ a . At point B, the second reflection and mode conversion occur, generating an ultrasonic shear wave. The second reflected wave passes through point A and arrives at the left ultrasonic transducer 17 at time τ b . Then the generated ultrasonic shear wave is reflected by the bottom surface of the specimen, converted into an ultrasonic surface wave at point B, passes through point A, and arrives at the left ultrasonic transducer 17 at time τ c . Therefore, the measured second time information includes τ a , τ b and τ c .

[0133] Step S1-8: Excite an ultrasonic surface wave through the left ultrasonic transducer 17, receive the reflected signal through the right ultrasonic transducer 18, and extract the third time information of the arrival of the reflected signal through the oscilloscope 15.

[0134] For surface cracks with a depth greater than 2λ, that is, the double-probe method is used for measurement, and the measured third time information is used to calculate the depth of the open crack.

[0135] In this embodiment, the left ultrasonic transducer 17 is placed at the left end of the roll, and the right ultrasonic transducer 18 is placed at the right end of the roll, and the defect and damage to be measured are within their measurement ranges. Other settings are the same as above. Then switch to circuit ③. The left ultrasonic transducer 17 is used to excite ultrasonic signals, the right ultrasonic transducer 18 is used to receive ultrasonic signals, and the oscilloscope 15 and the computer 16 are used to extract the arrival time τ of the reflected wave in the measurement signal d , τ e and τ f .

[0136] The ultrasonic surface wave excited by the left ultrasonic transducer 17 crosses the open crack and arrives at the right ultrasonic transducer 18 at time τ d . At the same time, the ultrasonic surface wave passes through point A, point B, and point C and arrives at time τ eArrive at the right ultrasonic transducer 18 at moment; the ultrasonic surface wave passes through point A, undergoes mode conversion at point B to generate an ultrasonic shear wave, reflects at the bottom surface of the specimen, converts to an ultrasonic surface wave at point B, passes through point C and arrives at the right ultrasonic transducer 18 at τ f Therefore, the measured third time information includes τ d 、τ e and τ f .

[0137] Repeat the above steps 5 times to obtain 30 pieces of time information.

[0138] For various types and degrees of roll surface defects or damages, use the above step S1 for measurement to obtain multiple groups of measurement information. These information will be used as the input of the BP neural network, and the corresponding defect or damage type and degree will be used as the output of the BP neural network.

[0139] Step S2, for different types and degrees of defects or damages, extract the measurement signal characteristics obtained in step S1.

[0140] Figure 1 is the schematic diagram of the roll damage classification and recognition method based on surface waves and neural networks in this embodiment.

[0141] As Figure 1 shown, for different types and degrees of defects or damages, analyze the measurement signal characteristics obtained in step S1.

[0142] The steel sticking on the roll surface will consume the energy of the ultrasonic surface wave, cause the attenuation of the ultrasonic surface wave, and make it unable to produce obvious reflections. Therefore, according to the change of the ultrasonic signal amplitude and the presence or absence of the reflection signal, it can be judged whether there is a steel sticking phenomenon. In this embodiment, the first amplitude signal and the second amplitude signal measured in step S1 are used for judgment.

[0143] For the buried defects or damages in the roll subsurface, in this embodiment, by changing the excitation frequency, the presence or absence and depth of the buried defects or damages are judged. From the distribution of displacement and stress along the thickness direction, it can be known that the energy of the ultrasonic surface wave is mainly concentrated in the surface layer with a thickness of 2λ and is sensitive to the buried defects or damages therein. Therefore, the detection depth of the ultrasonic surface wave is 2λ. According to it can be known that adjusting the frequency can change the detection depth of the ultrasonic surface wave. Therefore, in this embodiment, the third amplitude signal and the first time signal measured in step S1 are used for judgment.

[0144] During the process of adjusting the excitation frequency in the range of 100 kHz to 10.1 MHz, if the reflection signal received by the left ultrasonic transducer 17 has an excitation frequency of f xWhen it disappears, the defect or damage is a buried defect or buried damage, and the depth is The distance between it and the left ultrasonic transducer 17 is Calculated in the following way:

[0145] First, use the left ultrasonic transducer 17 to record the arrival time τ of the reflected wave g , then use the right ultrasonic transducer 18 to record the arrival time τ of the reflected wave h , the straight-line distance between the left ultrasonic transducer 17 and the right ultrasonic transducer 18 is D, and the distance y between the buried defect, buried damage and the left ultrasonic transducer 17 depends on:

[0146]

[0147] According to formula (8), it can be known that

[0148] During the above process of gradually increasing the excitation frequency, if the reflected signal never disappears, the corresponding defect or damage should be located on the surface of the roll, and the type of the defect or damage is surface crack, etc. The distance between the surface crack and the left ultrasonic transducer 17 is also The derivation process is the same as above.

[0149] For the depth of the surface crack, first, it is necessary to judge whether the depth of the surface crack is greater than the detection depth 2λ of the ultrasonic surface wave, and then for cracks with a depth less than or greater than 2λ, the accurate actual depth is obtained through different methods.

[0150] First, calculate a, b, c according to the Poisson's ratio of the roll material, select the measurement distance x and the reference crack length L, and verify the relationship diagram between the amplitude ratio Δ and the crack depth z / λ. Determine the minimum crack depth to be detected as 0.17λ, adjust the left ultrasonic transducer 17, at a distance x from the right angle side, adjust the height of the reflected wave to 80% max of the full screen, and record the gain G3 at this time. Refer to the distance y between the left ultrasonic transducer 17 and the ultrasonic surface wave, adjust the left ultrasonic transducer 17, at a distance x from the surface crack, adjust the height of the reflected wave to 80% max of the full screen, and record the gain G4 at this time. Obtain Δ Meas = G4 - G3, according to the relationship diagram between Δ and z / λ, calculate z Meas / λ. Measure the length of the surface crack and record it as the value L Meas . If L Meas is greater than L, then z Meas / λ is greater than the actual depth of the surface crack, otherwise, z Meas / λ is less than the actual depth of the surface crack. If Δ Meas is approximately equal to 0, then the depth of the surface crack is greater than 2λ.

[0151] Then, for cracks with a depth less than 2λ, in this embodiment, the depth of the surface crack is judged according to the difference Δ between the first gain and the second gain. The difference Δ between the second gain and the first gain is equal to:

[0152]

[0153]

[0154] In the formula, H B is the reflection wave height of the right-angle side in the oscilloscope, H f is the reflection wave height of the crack in the oscilloscope, x is the distance between the transmitting transducer and the crack or the right-angle side, L is the length of the measured crack, n is the ratio of the depth of the measured crack to the wavelength, and a, b, and c are related to the Poisson's ratio of the material. According to the calculated Δ, by referring to the relationship diagram between Δ and the crack depth z / λ, the depth of the surface crack can be determined.

[0155] For open cracks on the roll surface with a depth greater than 2λ, when the ultrasonic surface wave encounters an open crack, it propagates on its surface, reflects at the left end of the open crack (denoted as point A), and reflects and undergoes mode conversion at the tip of the open crack (denoted as point B). Based on the above physical phenomena, the single-probe method and the double-probe method are used to judge the depth of the open crack according to the time difference of the reflection waves.

[0156] In the single-probe method (i.e., the above steps S1-7), the left ultrasonic transducer 17 serves as both the excitation and receiving transducer at the same time. After the ultrasonic surface wave encounters an open crack, it undergoes the first reflection at point A and reaches the left ultrasonic transducer 17 at time τ a . The ultrasonic surface wave undergoes the second reflection and mode conversion at point B, generating an ultrasonic shear wave. Among them, the ultrasonic surface wave of the second reflection passes through point A and reaches the left ultrasonic transducer 17 at time τ b . In addition, the generated ultrasonic shear wave reflects at the bottom surface of the specimen, converts to an ultrasonic surface wave at point B, and then passes through point A and reaches the left ultrasonic transducer 17 at time τ c . Therefore:

[0157]

[0158]

[0159] In the formula, H is the thickness of the specimen, and h is the crack depth. And Both can be used to obtain the value of h. If τ c is difficult to identify or does not exist, then h only depends on formula (11).

[0160] In the dual-probe method (i.e., the above steps S1-8), the left ultrasonic transducer 17 serves as the excitation and receiving transducer, and the right ultrasonic transducer 18 serves as the receiving transducer. The ultrasonic surface wave crosses the open crack and directly reaches the right ultrasonic transducer 18 at time τ d At the same time, the ultrasonic surface wave travels along the surface of the open crack, passing through points A, B, and C (the right end of the open crack), and directly reaches the right ultrasonic transducer 18 at time τ e In addition, the ultrasonic surface wave passes through point A, undergoes mode conversion at point B to generate an ultrasonic shear wave. The generated ultrasonic shear wave is reflected from the bottom surface of the specimen, undergoes mode conversion again at point B to generate an ultrasonic surface wave that passes through point C and reaches the right ultrasonic transducer 18 at time τ f Therefore:

[0161]

[0162]

[0163] In the formula, and can corroborate each other. Similarly, when τ f is difficult to identify or does not exist, h can be obtained from formula (13).

[0164] Figure 4 is the flowchart of step S2 in this embodiment.

[0165] As Figure 4 shown, based on the above principle, step S2 specifically includes the following sub-steps:

[0166] Step S2-1: According to the first amplitude information, determine whether the maximum amplitude of the reflected signal in step S1-1 is greater than a predetermined value. If the determination is yes, proceed to step S2-4; if the determination is no, proceed to step S2-2.

[0167] Step S2-2: Further, according to the first amplitude signal and the second amplitude information, determine whether the decrease in the amplitude of the ultrasonic signal received by the right ultrasonic transducer 18 in step S1-2 exceeds a predetermined threshold. If the determination is yes, proceed to step S2-3.

[0168] Step S2-3: The type of surface defect or damage is steel bonding.

[0169] That is, in the above measurement, if the left ultrasonic transducer 17 does not receive an obvious reflected signal and the amplitude of the reflected signal received by the right ultrasonic transducer 18 decreases significantly, it is determined as steel bonding.

[0170] Step S2-3a: Compare with a non-destructive roll, and determine the size of the steel bonding on the roll surface according to the value of the second amplitude.

[0171] Step S2-4: When an obvious reflection signal is received, further determine whether the reflection signal disappears at a certain excitation frequency in Step S1-3. If the determination is yes, proceed to Step S2-5; if the determination is no, proceed to Step S2-6.

[0172] Step S2-5: Whether the reflection signal disappears at a certain excitation frequency, and the type of defect or damage is a buried defect or buried damage.

[0173] Step S2-5a: Calculate the depth of the buried defect or buried damage according to the excitation frequency when the reflection signal disappears.

[0174] Step S2-5b: Calculate the position information of the buried defect or damage according to the first time information.

[0175] As described above, the depth of the buried defect or damage is The distance between it and the left ultrasonic transducer is

[0176] Step S2-6: During the test in Step S1-3, the reflection signal never disappears, and the type of defect or damage is a surface crack.

[0177] Step S2-6a: Determine whether the surface crack is an open crack with a depth greater than 2λ. If the determination is no, proceed to Step S2-6b; if the determination is yes, proceed to Step S2-6c.

[0178] That is, according to the above method, determine whether the depth of the surface crack is greater than 2λ.

[0179] Step S2-6b: For a surface crack with a depth less than 2λ, calculate the difference Δ between the second gain and the first gain, and obtain the depth of the surface crack according to the gain difference Δ.

[0180] As described above, according to the calculated Δ, by referring to the relationship diagram between Δ and the crack depth z / λ, the depth of the surface crack can be determined.

[0181] Step S2-6c: For an open crack with a depth greater than 2λ, calculate the depth of the open crack according to the second time information or the third time information.

[0182] Using the single-probe method, the depth of the open crack is or Using the double-probe method, the depth of the open crack is or

[0183] Step S2-6d: Calculate the position information of the surface crack according to the first time information.

[0184] As described above, the distance between the surface crack and the left ultrasonic transducer is

[0185] Through the above steps, the specific type, depth, position information, etc. of the roll surface defects or damages corresponding to the measurement signals are analyzed and obtained.

[0186] Step S3: Combine the characteristics of the measurement signals, the corresponding surface defect or damage type and degree to form a training sample, and store the training sample in the training database.

[0187] That is, combine the measurement signal characteristics, the corresponding defect or damage type, depth, position information, etc. obtained in steps S1 and S2 to form a training sample, so as to construct a training database for the classification and recognition of roll surface defects and damages.

[0188] Step S4: Establish a BP neural network model, and use the training samples in the training database to train the BP neural network model to obtain a trained BP neural network model as the roll surface damage classification and recognition model.

[0189] In this embodiment, the method of machine learning is used to establish a BP neural network model with cognitive function for identifying steel sticking, indentation, surface cracks and buried defects or buried damages in the roll.

[0190] Figure 7 It is a schematic structural diagram of the BP neural network with cognitive function in this embodiment, that is, a schematic structural diagram of the roll surface damage classification and recognition model in this embodiment.

[0191] As Figure 7 shown, the input variables of the roll surface damage classification and recognition model include the time, frequency or amplitude information of multiple measurement signals:

[0192] Inp={Inp1,Inp2,L,Inp n} (15)

[0193] The above model belongs to a multi-layer forward feedback network and is obtained by training with a BP neural network. The BP neural network mainly includes an input layer, a hidden layer and an output layer. The activation functions f1 and f2 in the hidden layer and the output layer are mainly used to obtain the output variables Q and Outp.

[0194] The BP algorithm mainly includes two stages, namely the forward propagation stage and the backward propagation stage. In the forward propagation stage, the initial Inp signal is transmitted forward, from the input layer to the hidden layer, and the result Q is obtained:

[0195] Q=f1(W T Inp+B1) (16)

[0196] Then Q goes from the hidden layer to the output layer to obtain the training value Outp:

[0197] Outp = f2(V T Q + B2) (17)

[0198] Where W and V are the weight matrices between layers, B1 and B2 are the bias matrices, and f1 and f2 are activation functions, with the Sigmoid function used for activation:

[0199]

[0200] After the signal passes through forward propagation, the training values of each unit and the true value o i The expected value of the loss function is equal to:

[0201]

[0202] In the backpropagation stage, the above model continuously adjusts the weights of each layer according to the gradient descent idea to minimize the expected value of the loss function.

[0203] The training samples include the measurement signals obtained by the ultrasonic detection system 10 and the corresponding types and degrees of roll surface damage. The above model feeds back the output unit error and the hidden unit error to the BP neural network according to the training samples <Inp, Outp> to adjust the connection weights of the output unit parameters and the hidden unit parameters.

[0204] The error of the output unit is:

[0205]

[0206]

[0207] The error of the hidden unit is:

[0208]

[0209]

[0210] The output unit parameters are correspondingly updated to:

[0211]

[0212]

[0213] The hidden unit parameters are correspondingly updated to:

[0214]

[0215]

[0216] In the formula, η is the learning rate, k = 1, 2, …, n is the number of iterations, and ⊙ represents the Hadamard product.

[0217] In step S5, an acoustic measurement is performed on the surface of the roll using the ultrasonic measurement system 10 to obtain a measurement signal, and the measurement signal is input into the roll surface damage classification and recognition model, thereby obtaining the type and degree of defects or damages on the roll surface.

[0218] In this embodiment, there are five defects and damages distributed on the surface of the roll to be measured, including one type-1 damage, one type-2 damage, and three type-3 damages.

[0219] Before starting the measurement, set the initial position of the roll and mark it, adjust the clamping device 20 so that the two ultrasonic transducers are respectively located at both ends of the roll and pressed tightly, apply a coupling agent between the ultrasonic transducers and the roll, maintain the value of the pressure sensor at about 75 N, switch the circuits ①, ②, and ③ in the ultrasonic measurement system 10 to obtain the above-mentioned various measurement information, and identify the type, position, and degree of defects or damages in the initial position.

[0220] Then, drive the second servo motor 221 to drive the planetary gear train 214 to rotate, set a specific step size, and perform a circumferential scan on the roll. During the scan, identify the type and degree of defects or damages.

[0221] In this embodiment, after the circumferential scan and identification are completed, the identification results are also visualized. The information on the damage position, type, and degree is mapped onto the model of the outer surface of the roll, and a defect or damage distribution map of the roll surface is drawn, intuitively reflecting the evaluation results of the roll surface and subsurface.

[0222] Figure 8 It is a schematic diagram of the distribution of defects or damages on the roll surface and subsurface in this embodiment.

[0223] As Figure 8 shown, through the method of this embodiment, ultrasonic detection is performed on the roll surface, intelligent identification and classification of surface damages are carried out, and the identification and classification results are visualized. It can be seen that one type-1 damage, one type-2 damage, and three type-3 damages are accurately identified, and information such as the position, size, and depth of the defects and damages is obtained.

[0224] As described above, in this embodiment, defects or damages of different types and degrees on the roll surface are measured, a database of the above-mentioned measurement signals is constructed, the gain of the power amplifier, the amplitude, frequency, and time characteristics of the measurement signal are extracted, and combined with the type and degree of defects or damages, the BP neural network is trained. The trained BP neural network is used to identify the type and degree of roll defects or damages.

[0225] Figure 8 It is the structural block diagram of the roll damage classification and recognition device based on surface wave and neural network in this embodiment.

[0226] As Figure 8 shown, this embodiment also provides a roll damage classification and recognition device 100 based on surface wave and neural network, including an ultrasonic measurement system 110, a roll surface damage intelligent classification and recognition model 120, and a control unit 130 for controlling the work of both.

[0227] Among them, the structure and working principle of the ultrasonic measurement system 110 are the same as those of the above ultrasonic measurement system 10, and the roll surface damage intelligent classification and recognition model 120 is the same as the above roll surface damage intelligent classification and recognition model, so no repeated description will be given.

[0228] Function and effect of the embodiment

[0229] According to the roll damage classification and recognition method and device 100 based on surface wave and neural network provided in this embodiment, the ultrasonic measurement system 10 performs acoustic measurement on different types and degrees of defects or damages on the surface of the sample roll, and uses the measured signal characteristics, corresponding defect or damage types and degrees obtained from the analysis as training samples to construct a training database. Then, the data in the training database is used to train the BP neural network to obtain a roll surface damage classification and recognition model that can intelligently identify the defects and damages on the roll surface. Since ultrasonic surface wave is used to detect surface defect damages, the measured roll is not damaged, so the method of this embodiment is a non-destructive method; since the training data covers data of various different types and degrees of defects or damages, the trained model can well identify various defects and damages on the surface of the measured roll and give corresponding degree information.

[0230] Furthermore, the ultrasonic measurement system 10 includes a left ultrasonic transducer 17, a right ultrasonic transducer 18, and a clamping device 20 for clamping and positioning the two ultrasonic transducers. The clamping device 20 includes a circumferential positioning module 21, an axial driving module 22, and a radial pressing module 23. Therefore, the two ultrasonic transducers can be respectively positioned at both ends of the roll and pressed against the roll, and the two ultrasonic transducers can be rotated along the circumference of the roll, so that the roll can be scanned circumferentially to identify all defects and damages distributed on its surface.

[0231] Furthermore, the ultrasonic measurement system 10 includes an arbitrary waveform generator 11, a power amplifier 12, a duplexer 13, filters 14a and 14b, an oscilloscope 15, and a computer 16. The left ultrasonic transducer 17 and the right ultrasonic transducer 18 are respectively connected to the two filters 14a and 14b, forming three lines. Therefore, different ultrasonic measurement methods can be adopted through the three lines to obtain various information such as amplitude, time, and frequency in the measurement information. Furthermore, information such as the type, degree, and position of surface defects or damages can be calculated based on this information and the principle of ultrasonic measurement, providing high-precision training samples, so that the trained model can achieve a higher-precision classification and recognition effect.

[0232] The above embodiments are only used to illustrate the specific implementation manners of the present invention, and the present invention is not limited to the description scope of the above embodiments.

Claims

1. A method for classifying and identifying roll damage based on surface waves and neural networks, characterized in that, It includes the following steps: Step S1: Using an ultrasonic measurement system, generate ultrasonic surface waves to acoustically measure the defects or damages on the surface of the sample roll, and obtain a measurement signal; Step S2: Extract the characteristics of the measurement signal for different types and degrees of the defects or damages; Step S3: Compose the characteristics of the measurement signal, the corresponding types and degrees of the defects or damages into a training sample, and store it in a training database; Step S4: Establish a BP neural network model, and use the training database to train the BP neural network model to obtain the trained BP neural network model as a classification and recognition model for roll surface damage; Step S5: Use the ultrasonic measurement system to acoustically measure the roll surface, collect the measurement signal, and input the measurement signal into the classification and recognition model for roll surface damage, so as to obtain the types and degrees of the defects or damages on the roll surface Among them, the ultrasonic measurement system includes an arbitrary waveform generator, a power amplifier, a left ultrasonic transducer, a right ultrasonic transducer, a duplexer, an oscilloscope, a first filter, a second filter, a computer, and an ultrasonic transducer clamping device, The arbitrary waveform generator is connected to the power amplifier and the oscilloscope, The power amplifier is connected to the duplexer, The duplexer is connected to the first filter and the oscilloscope, The oscilloscope is also connected to the second filter and the computer, The left ultrasonic transducer is connected to the first filter, The right ultrasonic transducer is connected to the first filter and the second filter, Both the left ultrasonic transducer and the right ultrasonic transducer are installed on the ultrasonic transducer clamping device, Step S1 includes the following sub-steps: Step S1-1: Install the sample roll in the ultrasonic transducer clamping device, and adjust the positions of the left ultrasonic transducer and the right ultrasonic transducer so that the defect or damage is within its detection area. Excite ultrasonic surface waves and receive reflected signals through the left ultrasonic transducer at the initial excitation frequency, and extract the first amplitude information of the reflected signal through the oscilloscope; Step S1-2: Excite ultrasonic surface waves through the left ultrasonic transducer at the initial excitation frequency, receive ultrasonic signals through the right ultrasonic transducer, and extract the second amplitude information of the reflected signal through the oscilloscope; Step S1-3: Excite ultrasonic surface waves and receive reflected signals through the left ultrasonic transducer, and adjust the excitation frequency of the ultrasonic surface waves through the arbitrary waveform generator, gradually increasing from the first excitation frequency to the second excitation frequency at a predetermined step length, and extract the third amplitude information of the reflected signal through the oscilloscope during this process; Step S1-4: Excite ultrasonic surface waves and receive reflected signals through the left ultrasonic transducer and the right ultrasonic transducer in sequence, and extract the first arrival time information of the reflected signal through the oscilloscope Step S1-5: Adjust the position of the left ultrasonic transducer so that its distance from the right end of the roll is a predetermined distance. Excite the ultrasonic surface wave through the left ultrasonic transducer and receive the reflected signal. Observe the waveform of the reflected signal on the oscilloscope, and adjust the height of the waveform to a predetermined height through the power amplifier. Record the first gain of the power amplifier at this time. Step S1-6: Adjust the position of the left ultrasonic transducer so that its distance from the defect or damage is the predetermined distance. Excite the ultrasonic surface wave through the left ultrasonic transducer and receive the reflected signal. Observe the waveform of the reflected signal on the oscilloscope, and adjust the height of the waveform to a predetermined height through the power amplifier. Record the second gain of the power amplifier at this time. Step S1-7: Adjust the positions of the left ultrasonic transducer and the right ultrasonic transducer so that the defect or damage is within their detection area. Excite the ultrasonic surface wave through the left ultrasonic transducer and receive the reflected signal, and extract the second time information when the reflected signal arrives through the oscilloscope. Step S1-8: Excite the ultrasonic surface wave through the left ultrasonic transducer, receive the reflected signal through the right ultrasonic transducer, and extract the third time information when the reflected signal arrives through the oscilloscope.

2. The method for classifying and identifying roll damage based on surface waves and neural networks according to claim 1, It is characterized in that: Wherein, the ultrasonic transducer clamping device includes: A pair of circumferential positioning modules, respectively arranged on both sides of the roll, each having a turntable; An axial driving module, arranged on the turntables of the pair of circumferential positioning modules, capable of rotating along the outer periphery of the roll; and A pair of radial pressing modules; movably arranged on the axial driving module, The left ultrasonic transducer and the right ultrasonic transducer are respectively installed on a pair of the radial pressing modules, and can be driven by the radial pressing modules to press towards the roll surface.

3. The method for classifying and identifying roll damage based on surface wave and neural network according to claim 1, characterized in that: Among them, The initial excitation frequency is 1 MHz, The first excitation frequency is 100 kHz, the second excitation frequency is 10.1 MHz, and the predetermined step size is 0.5 MHz to 2.5 MHz, The predetermined height is 80% of the full screen height of the oscilloscope, The first time information includes the arrival time τ of the reflected signal at the left ultrasonic transducer g and the arrival time τ of the reflected signal at the right ultrasonic transducer h .

4. The method for classifying and identifying roll damage based on surface wave and neural network according to claim 1, characterized in that: Among them, The left end of the surface crack is point A, the bottom tip is point B, and the right end is point C, In step S1-4, the first reflection of the surface acoustic wave occurs at point A, and the reflected wave reaches the left ultrasonic transducer at time τ a At time a , the second reflection and mode conversion occur at point B to generate a surface acoustic wave. The reflected wave of the second reflection passes through point A and reaches the left ultrasonic transducer at time τ b At time b , the generated surface acoustic wave is reflected at the bottom surface, converted to the surface acoustic wave at point B, passes through point A again, and reaches the left ultrasonic transducer at time τ c At time c . The second time information includes τ a , τ b , τ c , In step S1-7, the surface acoustic wave crosses the surface crack and reaches the right ultrasonic transducer at time τ d At the same time, the surface acoustic wave passes through point A, point B, and point C and reaches the right ultrasonic transducer at time τ e At this time The ultrasonic surface wave passes through point A, undergoes mode conversion at point B to generate an ultrasonic shear wave, reflects from the bottom surface, is converted back to an ultrasonic surface wave at point B, passes through point C and reaches the right ultrasonic transducer at time τ f ​ The third time information includes τ d , τ e , τ f .

5. The method for classifying and identifying roll damage based on surface waves and neural networks according to claim 4, It is characterized in that: Wherein, step S2 includes the following sub-steps: Step S2-1: According to the first amplitude information, judge whether the maximum amplitude of the reflected signal in step S1-1 is greater than a predetermined value; Step S2-2: If the judgment in step S2-1 is no, further judge whether the decrease amount of the amplitude of the ultrasonic signal received by the right ultrasonic transducer in step S1-2 exceeds a predetermined threshold according to the first amplitude information and the second amplitude information; Step S2-3: If the judgment in step S2-2 is yes, the type of the defect or damage is steel sticking. Step S2-4, if the judgment in step S2-1 is yes, further judge whether the reflected signal disappears at several of the excitation frequencies in step S1-3; Step S2-5, if the judgment in step S2-4 is yes, the type of the defect or damage is a buried defect or a buried damage; Step S2-6, if the judgment in step S2-4 is no, the type of the defect or damage is a surface crack.

6. The method for classifying and identifying roll damage based on surface waves and neural networks according to claim 5, It is characterized in that: Wherein, step S2 further includes the following sub-steps: Step S2-5a, calculate the depth of the buried defect or the buried damage according to the excitation frequency when the reflected signal disappears; Step S2-5b, calculate the position information of the buried defect or the buried damage according to the first time information; Step S2-6a, judge whether the surface crack is an open crack; Step S2-6b, if the judgment in step S2-6a is no, calculate the amplitude ratio of the reflected signal according to the difference ∆ between the second gain and the first gain, and obtain the depth and position information of the surface crack according to the amplitude ratio; Step S2-6c, if the judgment in step S2-6a is yes, calculate the depth of the surface crack according to the second time information or the third time information; Step S2-6d, calculate the position information of the surface crack according to the first time information.

7. The method for classifying and identifying roll damage based on surface waves and neural networks according to claim 6, characterized in that: Among them, In step S2-5a, the depth of the buried defect or the buried damage is , In step S2-5b, the distance between the buried defect or the buried damage and the left ultrasonic transducer is , Where c R is the velocity of the ultrasonic surface wave in the sample roll, f x is the excitation frequency when the reflected signal disappears in step S1-3, and D is the distance between the left ultrasonic transducer and the right ultrasonic transducer. In step S2-6b, the difference ∆ between the second gain and the first gain is: ， ， Wherein, H B is the height of the reflected wave of the right-angled side in the oscilloscope, H f is the height of the reflected wave of the surface crack in the oscilloscope, x is the distance between the left ultrasonic transducer and the surface crack or the right-angled side, L is the length of the surface crack, n is the ratio of the depth of the surface crack to the wavelength, a, b, c are related to the Poisson's ratio of the material of the roll, and 2λ is the detection depth of the ultrasonic surface wave. In step S2-6c, the depth of the surface crack is or , or is or , where c T is the velocity of the ultrasonic shear wave in the sample roll, In step S2-6d, the distance between the surface crack and the left ultrasonic transducer is .

8. A roll damage classification and recognition device based on surface waves and neural networks, characterized in that, Including: An ultrasonic measurement system for acoustically measuring the surface of the roll to obtain a measurement signal; And A roll surface damage classification and identification model for classifying and identifying the measurement signal to obtain the corresponding type and degree of the defect or damage on the roll surface, Wherein, the roll surface damage classification and identification model is a trained BP neural network model, and its training samples are obtained through the following steps: Use the ultrasonic measurement system to acoustically measure the defects or damages on the surface of the sample roll to obtain a measurement signal; Extract the characteristics of the measurement signal for different types and degrees of defects or damages; Form the training samples with the characteristics of the measurement signal, the corresponding type and degree of the defect or damage; Wherein, the roll surface damage classification and identification device based on surface waves and neural networks uses the method for classifying and identifying the type and degree of the defect or damage on the roll surface according to any one of claims 1-7.

Images