Manually controlled injection pen

Abstract

The manual injection acoustic precision control metering method comprises the following steps: rotating the main shell, the ratchet groove collides with the ratchet teeth on the ratchet wheel, the switch sensor and the perception sensor obtain the collision sound and vibration signal, the signal detected by the sound sensor is classified by the MFCC algorithm and DNN, the sound characteristics of the processing and recognition are output, the signal detected by the vibration sensor is classified by the EMD algorithm and CDD, the vibration signal of the processing and recognition is output, and finally the output sound characteristics, the vibration signal and the time difference between the received sound signal and the vibration signal are input into the SVM classifier for final recognition and classification, and whether the manual injection is performed correctly is identified, so that the manual injection can be accurately recorded. The algorithm design for classifying and identifying sound and vibration of the present invention is scientific, and the sound and vibration signals generated during manual injection are intelligently collected and recognized, so that the recording of insulin injection amount is more accurate, and it is more convenient to use and has better universality.

Description

Technical Field

[0001] The invention belongs to the technical field of insulin injection metering, and in particular relates to a manually controlled injection pen. Background Art

[0002] Insulin injection devices are devices that mimic the secretion of insulin in normal human bodies and infuse insulin into unhealthy human bodies. Diabetic patients need to carry their insulin with them even when exercising and sleeping, except for special circumstances such as bathing, in order to effectively regulate blood sugar. In the development of insulin injection devices, syringes, insulin jet syringes, insulin pumps, insulin injection pens, etc. have appeared one after another. Syringes are traditional insulin injection devices, and their volumes vary, and the needles of syringes also have different lengths and diameters. The shortcomings of this injection device are that patients must have good eyesight and flexible hands; it is inconvenient to carry it when going out; and the skin pain caused by the injection needle. The insulin pen is a pen-type insulin injection device, which is a revolutionary turning point in the field of diabetes treatment. The appearance is similar to a fountain pen, but the size is slightly larger, which can reduce the psychological burden of insulin injection patients. Its internal structure is ingenious, combining insulin and injection devices into one.

[0003] Compared with ordinary syringes, insulin injection pens have the following advantages: (1) Insulin and injection device are combined into one, eliminating the tedious process of extracting insulin and making it easy to replace the refill; (2) It is easy to carry; (3) The operation is simple and flexible; (4) The injection process is simpler and more concealed; (5) The dosage is more accurate, with a minimum infusion volume of 0.1u unit; (6) Less pain.

[0004] There are generally two injection methods for insulin pens: one is manual injection, which is to manually inject insulin into one's body, which requires the injector to concentrate to prevent excessive or insufficient injection, which is more cumbersome; the other method is automatic injection, which relies on related devices and sound and light reminders to complete the insulin injection, and cannot be used normally when there is no electricity in the wild.

[0005] When manual insulin injection is used without electricity, the amount of insulin injected will be confused during daily injection. Manually injecting insulin into the body requires the injector to concentrate to prevent excessive or insufficient injection, which is cumbersome. Therefore, the daily measurement method for manual insulin injection is very important.

[0006] The Chinese invention patent (application number: 202111070240.5, publication number: CN 113730729A) discloses a manual-electric integrated insulin injection pen and its injection method, which still has the following defects when used: the manual injection metering cycle is difficult to accurately determine during personal operation, and the size of the collision sound between the ratchet and the ratchet groove is different (affected by factors such as the depth of the ratchet groove, the service life of the insulin injection pen, the season of insulin injection, and the ambient temperature of insulin injection), so the accuracy is still not high enough during manual injection operation. In addition, during manual injection, you need to concentrate on recording the number of collision sounds between the ratchet and the ratchet groove (the amount of injection can be known based on the reminder of the sound), which is easy to cause counting errors for the elderly who operate by themselves. Summary of the invention

[0007] In order to solve the deficiencies in the prior art, the present invention provides a manually controlled injection pen which has high injection volume accuracy, automatic metering and wide universality during manual injection.

[0008] In order to solve the above technical problems, the present invention adopts the following technical solution: a manual injection acoustic precision control metering method, comprising the following steps:

[0009] A. When the power of the injection pen is less than 20%, it will enter the energy-saving mode and can only be operated manually. When the user picks up the injection pen, the sensing sensor set on the left side of the inner wall of the main shell sends a signal to the controller, and the controller receives the signal that someone wants to use the injection pen;

[0010] B. Remove the pen cap and needle cap, and then push the metering ring to the left to separate the metering ring from the left end of the main shell, so that each card block is disengaged from each card slot, and the triangular ridges on the left side of each elastic card plate move to the left to the right side of the corresponding limit sector plate. At this time, the main shell rotates relative to the second connecting sleeve and the pen core; the switch sensor arranged on the left side of the inner wall of the main shell detects that the metering ring moves to the left, and transmits a signal to the controller. The controller receives the signal, and the manual injection mode is turned on;

[0011] C. The user inserts the needle into the human body, holds the refill sleeve and metering ring with one hand, and holds the main housing with the other hand, and rotates the main housing, which drives the internal gear box to rotate. The power output shaft of the gear box is in a locked state at this time, and the power output shaft of the gear box drives the screw to rotate. The screw drives the threaded hollow push rod to move left, and the hollow push rod pushes the injection piston to move left in the refill. The injection piston injects the insulin liquid to be injected in the refill into the human body;

[0012] D. During the rotation of the main housing, the anti-reverse rotation of the inner wall of the main housing rotates relative to the ratchet wheel, and the ratchet groove on the inner circle of the anti-reverse rotation and the ratchet teeth on the outer circle of the ratchet wheel will make a collision sound and vibrate at the same time. A sound sensor and a vibration sensor are arranged inside the ratchet wheel. Each collision sound signal is collected by the sound sensor, and each vibration signal is collected by the vibration sensor. The sound sensor and the vibration sensor transmit the collected signals to the controller. The controller classifies and identifies the sound and vibration signals through the internal algorithm, records and stores the injection amount of insulin liquid. Each successful identification represents the injection of 0.5u of insulin liquid, so as to achieve accurate calculation of the injection amount during manual injection; the injection amount is displayed on the display screen, and the user only needs to observe the value on the display screen to know the injection amount;

[0013] E. When the target injection volume is reached, stop rotating the main housing and move the metering ring to the right to reset it, so that each card block is correspondingly connected to each card slot, and the main housing and the pen core can be kept fixed; finally, close the needle cap and the pen cap.

[0014] The inner circle of the non-reversal wheel is evenly provided with 9 ratchet grooves along the circumferential direction, and the central angle between two adjacent ratchet grooves is 40°. The 9 ratchet grooves are divided into three groups of the same order, namely group 1, group 2 and group 3. Three adjacent ratchet grooves form a group, and the depths of the three ratchet grooves in each group are different, increasing in sequence clockwise. The depth layout of the three groups of ratchet grooves is consistent. Three ratchet teeth on the outer circle of the ratchet wheel are evenly provided along the circumferential direction, and the central angle between two adjacent ratchet teeth is 120°. When the ratchet wheel rotates, the three ratchet teeth collide with the ratchet grooves of the same depth at the same time. When the ratchet groove rotates 40°, the sound and vibration emitted when the ratchet grooves collide with the ratchet teeth are different; in step D, the internal algorithm of the controller includes feature processing of the sound signal and the vibration signal when the ratchet groove on the inner circle of the non-reversal wheel collides with the ratchet teeth on the outer circle of the ratchet wheel, and after the internal algorithm processing of the controller and the classification and recognition by the neural network, the sound signal recognition result, the vibration signal recognition result and the detection time difference T between the sound signal and the vibration signal are uploaded to the support vector machine (SVM) for classification and recognition again.

[0015] The characteristic processing method of the sound signal when the ratchet groove on the inner circle of the anti-reversal wheel collides with the ratchet teeth on the outer circle of the ratchet wheel adopts the Mel reciprocal coefficient (MFCC) method. MFCC is extracted based on Mel frequency. The mapping relationship between Mel frequency and normal Hertz frequency adopts formula (1):

[0016]

[0017] Where f Mel is the Mel frequency, the unit is MEL, f is the linear frequency, the unit is Hz;

[0018] The specific process of the Mel Reciprocal Coefficient (MFCC) method includes the following steps:

[0019] 1) Collecting sound signals: When the ratchet groove collides with the ratchet teeth on the outer circle of the ratchet wheel, the sound of the collision is transmitted to the controller through the sound sensor set inside the ratchet wheel. The sound signal collection time for each time is 0.2-0.5s;

[0020] 2) Sound preprocessing: The preprocessing process mainly involves pre-emphasis, framing and windowing of sound signals;

[0021] The pre-emphasis of the sound signal is usually performed by filtering. Here, the input signal x(n) is passed through a high-pass filter. The filter is V(n) = x(n) - a1x(n-1) in the time domain and H(z) = 1 - a1z in the frequency domain. -1 , where 0.9≤a1≤1.0, a1 is the pre-emphasis coefficient and is preferably 0.97;

[0022] When dividing the frame, the "short-term stable characteristics" of the sound signal are used to decompose the signal into several frames. Each frame can be set to 20ms, and the previous and next frames overlap by half. After dividing the frame, the time domain signal x(n) is combined into x i (n), i represents the relative frame number;

[0023] Windowing is to multiply the original function and the window function; let x(n) be the time domain signal and w(n) be the window function, then the N-point sequence x obtained by truncating x(n) with w(n) is n (n) See formula (2):

[0024] x n (n) = w(n)x(n) (2)

[0025] This patent selects the Hamming window, and formula (3) is as follows:

[0026]

[0027] Where W(n, a2) represents the window function, n=1,2,3,…,N is the sampling point number of the window function, and a2 represents the middle position of the window, preferably 0.46;

[0028] 3) Fast Fourier Transform (FFT): for x i (n) After FFT, we get X i (K), by combining formula (2) with discrete Fourier transform (DFT) calculation formula (4), we can get X i (K) is calculated by formula (5), where the Hamming window is used:

[0029]

[0030]

[0031] In formula (4), x(n) is the original signal, X N (k) is the coefficient after DCT transformation, and N is the number of points of the original signal;

[0032] Formula (5) N is the length of the DFT transform, k is the frequency, and h(n) is a window function with the same length as the sample;

[0033] 4) Mel filter bank: The Mel frequency filter bank is a group of triangular bandpass filters H arranged regularly within the specified frequency spectrum of the acoustic signal. m (k), 0≤m≤M, M is the number of filters, the number of filters is usually between 24 and 40, the center frequency of each triangular filter is f(m), and the transfer function formula (6) is as follows:

[0034]

[0035] Where H m (k) is the frequency response of the Mel filter, and the preferred number of M is 24;

[0036] 5) Obtaining logarithmic energy: In order to make the Mel spectrum obtained by the Mel filter bank more robust to noise and spectrum estimation errors, the logarithmic energy of the Mel spectrum is often obtained. The transfer function formula (7) from the linear spectrum to the logarithmic spectrum is as follows:

[0037]

[0038] Where X(k) is the linear spectrum, S(m) is the logarithmic spectrum, and m is the number of triangular bandpass filters;

[0039] 6) Discrete cosine transform: The logarithmic spectrum S(m) is subjected to discrete cosine transform (DCT) and can be converted back to the time domain. At this time, the time domain is the cepstrum domain, and the Mel frequency cepstrum coefficient can be obtained. The calculation formula (8) is as follows:

[0040]

[0041] Where C(b) represents the MFCC of the bth filter, L is the dimension of the extracted feature parameter MFCC; M is the number of Mel filter groups; the above is the process of extracting Mel frequency cepstral coefficients (MFCC);

[0042] 7) Fully connected neural network (DNN) classification: The MFCC static characteristics of acoustic detection are input into the fully connected neural network (DNN) model, and the recognition results are output. The signals finally processed and recognized by the sound signals are recorded as voice 1, voice 2, voice 3 and voice 4 respectively; voice 1 is the sound coefficient generated by the collision between the ratchet groove with the largest depth and the ratchet, voice 2 is the sound coefficient generated by the collision between the ratchet groove with the second largest depth and the ratchet, voice 3 is the sound coefficient generated by the collision between the ratchet groove with the smallest depth and the ratchet, and voice 4 is the sound coefficient generated by other external collisions; sample training is performed based on the fully connected neural network, and the sample factors include the depth of the ratchet groove, the service life of the insulin syringe, the season of insulin injection and the ambient temperature of insulin injection. 2000 groups of samples are trained, and 500 groups of each of voice 1, voice 2, voice 3 and voice 4 are trained; the fully connected neural network (DNN) model is a conventional model and will not be repeated here.

[0043] The special diagnosis and processing method of the vibration signal when the ratchet groove on the inner circle of the anti-reversal wheel collides with the ratchet teeth on the outer circle of the ratchet wheel adopts the empirical mode decomposition method (EMD). EMD uses the time scale characteristics of the signal to decompose the signal without setting the basis function. It is essentially a time-frequency analysis method, which mainly completes the decomposition of a non-stationary signal. After decomposition, the linear sum of multiple IFMs and a trend term is obtained, that is, a non-stationary signal y(t) can be expressed as:

[0044]

[0045] Among them, IMF i (t) represents the i-th order intrinsic mode function, r(t) is the low-frequency pulsation, i.e., the trend term;

[0046] The specific process of the empirical mode decomposition (EMD) method includes performing EMD and convolutional neural network (CNN) classification on a vibration signal in sequence.

[0047] The specific steps for performing EMD on a signal are:

[0048] ① Collect vibration signals: When the ratchet groove collides with the ratchet teeth on the outer circle of the ratchet wheel, the vibration signal of the collision is transmitted to the controller through the vibration sensor set inside the ratchet wheel. The collection time of each vibration signal is 0.2-0.5s;

[0049] ② Find all the maximum and minimum points of the vibration signal y(t), and then connect all the maximum and minimum points with a curve to obtain the upper envelope e max (t) and the lower envelope e min(t), and then find the mean envelope function d(t) of the upper envelope and the lower envelope:

[0050]

[0051] ③ In formula (10), the difference between y(t) and d(t) is h1(t). If h1(t) is not an IMF, continue the above process until the kth cycle h 1k (t) is an IMF, denoted as c1(t);

[0052] ④According to the decomposed IMF, the residual signal y1(t) is obtained:

[0053] y1(t)=y(t)-c1(t) (11)

[0054] ⑤ Repeat the above steps until the nth order h n (t) becomes a monotonic sequence, and the trend term r is defined n (t) is y n (t).

[0055] The specific process of convolutional neural network (CNN) classification is as follows: the vibration preprocessing signal of vibration detection is input into the convolutional neural network (CNN) model, and the recognition result is output; the signals finally processed and recognized by the vibration signal are recorded as vibration 1, vibration 2, vibration 3 and vibration 4. Vibration 1 is the vibration processing signal generated by the collision between the ratchet groove with the largest depth and the ratchet wheel, vibration 2 is the vibration processing signal generated by the collision between the ratchet groove with the second largest depth and the ratchet wheel, vibration 3 is the vibration processing signal generated by the collision between the ratchet groove with the largest depth and the ratchet wheel, and vibration 4 is the vibration processing signal of other external collisions; sample training is performed based on the fully connected neural network, and the sample factors include the depth of the ratchet groove, the service life of the insulin syringe, the season of insulin injection and the ambient temperature of insulin injection, etc., and 2000 groups of samples are trained, and 500 groups of samples are trained for each type of vibration 1, vibration 2, vibration 3 and vibration 4; the convolutional neural network (CNN) model is a conventional model and will not be repeated here.

[0056] The sound signal recognition result, the vibration signal recognition result and the detection time difference T between the sound signal and the vibration signal are uploaded to the support vector machine (SVM) for classification and recognition again as follows:

[0057] There are many kinds of sound signals and vibration signals in the process of manual insulin injection, so this is a multi-classification problem. In the process of sound detection and vibration detection, the fundamental frequency periods of sound signals and vibration signals are different, so we need to train SVM model samples according to actual conditions. Sample training is based on the SVM model. The sample factors include the depth of the spine groove, the service life of the insulin syringe, the season of insulin injection, the detection time difference T of the sound signal and the vibration signal, and the ambient temperature of insulin injection. 2000 groups of samples are trained;

[0058] The voice 1, voice 2, voice 3, voice 4 finally processed and identified by the sound signal and the vibration 1, vibration 2, vibration 3, vibration 4 finally processed and identified by the vibration signal are input into the SVM model, and then the detection time difference T between the sound signal and the vibration signal is also input into the SVM model. The output results are: normal injection information, other syringe injection information and invalid injection information through SVM model recognition; this can ensure the accuracy of manual injection of insulin injection;

[0059] There are two types of manual injection metering cycles. The first type is the entire time period from the switch sensor being turned on to being turned off. The second type is after the switch sensor receives the signal, the manual injection device may not turn off the manual injection mode due to other reasons. At this time, when the sensor does not sense any external contact for 10 minutes, the metering is automatically terminated. The amount of insulin injected each time is shown in Formula 12:

[0060] S=0.5b1+b2×0 (12)

[0061] Where S is the injection volume per injection cycle, b1 is the number of injection device detection and identification times, and b2 is the number of invalid detection and identification times;

[0062] The first situation is as follows: when a user uses an insulin injection device for manual injection, he removes the pen cap and needle cap, and then pushes the metering ring to the left to separate the metering ring from the left end of the main shell. At this time, the switch sensor transmits a signal to start manual injection, and the injection cycle begins. After the patient completes the manual injection, he pushes the metering ring to the right to fit the metering ring with the left end of the main shell. At this time, the switch sensor transmits a signal to close manual injection, and the injection cycle ends.

[0063] The second type of situation is as follows: when a patient is using an insulin injection device for manual injection, he removes the pen cap and needle cap, and then pushes the metering ring to the left to separate the metering ring from the left end of the main shell. At this time, the switch sensor transmits a signal to start manual injection, and the injection cycle begins. When something unexpected happens to the patient during the injection, the manual injection is not completed, and the manual injection mode is not turned off. At this time, the sensing sensor senses that the patient has not touched the injection device within 10 minutes, and the metering cycle is automatically ended.

[0064] By adopting the above technical solution, when manual injection of insulin is required, the switch sensor and the perception sensor can detect the status of the injection device, the signal detected by the sound sensor is classified through the MFCC algorithm and DNN, and the sound characteristics of the output processing and recognition are output, the signal detected by the vibration sensor is classified through the EMD algorithm and CDD, and the vibration signal of the output processing and recognition is output, and finally the output sound characteristics, vibration signal and the time difference between the received sound signal and the vibration signal are input into the SVM classifier for final recognition and classification to identify whether the manual injection is performed correctly, so that the manual injection can be accurately recorded.

[0065] The algorithm for classifying and identifying sound and vibration in the present invention is scientifically designed, and the sound and vibration signals generated during manual injection are intelligently collected and identified, making the recording of insulin injection volume more accurate, more convenient to use, and more universal. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 is a structural cross-sectional view of the injection pen of the present invention;

[0067] Figure 2 This is an axonometric diagram of the overall structure of the injection pen after removing the main shell and metering ring;

[0068] Figure 3 is an axonometric view of the second connecting sleeve in the injection pen;

[0069] Figure 4 Schematic diagram of the connection between the injection pen stop reversal and the ratchet in the main housing;

[0070] Figure 5 is a schematic diagram of the shape of the ratchet groove in the injection pen;

[0071] Figure 6 It is a flow chart of the algorithm processing of sound signals of the present invention;

[0072] Figure 7 It is a flow chart of the vibration signal algorithm processing of the present invention;

[0073] Figure 8 This is a flow chart of SVM classification and recognition according to the present invention.

[0074] Figure 1-5 The reference numerals in the figure are respectively: pen cap 1, pen core 2, needle 3, pen core sleeve 4, main shell 5, controller 6, micro battery life battery, metering ring 8, needle cap 9, injection piston 10, electric motor 11, gear box 12, screw 13, hollow push rod 14, pipe column 15, switch button 16, first connecting sleeve 17, second connecting sleeve 18, limit ring plate 19, limit cylinder 20, card groove 21, annular boss 22, card block 23, limit fan-shaped plate 24, sliding through hole 25, elastic card plate 26, triangular convex ridge 27, anti-reverse 28, ratchet 29, ratchet groove 30, plastic column 31, display screen 32, sound sensor 33, vibration sensor 34, switch sensor 35, perception sensor 36. DETAILED DESCRIPTION

[0075] like Figure 1-Figure 5 As shown, the manually controlled injection pen in the present invention comprises a pen cap 1, a pen core 2, a needle 3, a pen core sleeve 4, a main shell 5, a power assembly, a controller 6, a micro battery 7 and a metering ring 8. The pen cap 1, the pen core 2, the needle 3, the pen core sleeve 4 and the main shell 5 are co-centered and horizontally arranged in the left-right direction. The right end of the pen cap 1, the right end of the pen core 2, the right end of the pen core sleeve 4 and the left end of the main shell 5 are all open. The pen core sleeve 4 is fixedly sleeved on the outside of the pen core 2, the pen cap 1 is pluggable and sleeved on the outside of the pen core sleeve 4, the needle 3 is fixedly installed at the center of the left end of the pen core sleeve 4, and the right end of the needle 3 passes through the pen core sleeve 4 and The left end of the refill 2 is connected to the interior of the refill 2, the left end of the needle 3 is located on the left side of the refill sleeve 4, the left end of the refill sleeve 4 is plugged and connected with a needle cap 9 sleeved on the outside of the needle 3, the right end of the refill 2 is connected to the left end of the main shell 5 through a connecting assembly, an injection piston 10 is sealingly and slidably arranged inside the refill 2, the power assembly, the controller 6 and the micro battery life battery 7 are all installed in the main shell 5, the metering ring 8 is slidably sleeved on the outside of the connecting assembly, the power assembly drives the injection piston 10 to move to the left, the micro battery life battery 7 is electrically connected to the power assembly and the controller 6 respectively, and the controller 6 is signal-connected to the power assembly.

[0076] The power assembly includes an electric motor 11, a gear box 12, a screw 13 and a hollow push rod 14. The electric motor 11 is fixedly installed in the right side of the main shell 5, the gear box 12 is fixedly installed in the middle of the main shell 5, the outer circumference of the gear box 12 is clamped and fixed with the inner wall of the main shell 5, the screw 13 is coaxially rotatable and arranged in the left side of the main shell 5, the middle part of the left end face of the gear box 12 is integrally formed with a column 15, the right end of the screw 13 coaxially extends into the column 15 and is coaxially connected to the left end of the power output shaft in the center of the gear box 12 as a whole, the left end of the screw 13 coaxially passes through the connecting assembly and is located on the right side of the right end of the pen core 2, The hollow push rod 14 is concentrically sleeved on the outside of the screw 13, and the right end portion of the inner circle of the hollow push rod 14 is provided with an internal thread that cooperates with the external thread of the screw 13. The left end of the hollow push rod 14 extends into the refill 2 and is fixedly connected to the right end face of the injection piston 10. The power shaft of the electric motor 11 is horizontally arranged in the left and right directions, and the left end of the power shaft of the electric motor 11 is transmission-connected to the power input shaft of the gear box 12. The micro battery 7 is electrically connected to the electric motor 11, and the controller 6 is signal-connected to the electric motor 11. A switch button 16 for controlling the operation of the electric motor 11 is provided in the middle of the right side surface of the main shell 5.

[0077] The connecting assembly includes a first connecting sleeve 17 and a second connecting sleeve 18. The first connecting sleeve 17 is fixedly sleeved on the outside of the second connecting sleeve 18 with the same center. The left end of the first connecting sleeve 17 is pressed against the right end of the refill 2. The right end of the refill sleeve 4 is sleeved on the left end of the first connecting sleeve 17 and threaded together. The second connecting sleeve 18 is sleeved on the outside of the hollow push rod 14 with the same center. The right end of the second connecting sleeve 18 extends into the left side of the main shell 5 with the same center. The outer circle of the right end of the second connecting sleeve 18 is integrally formed with a limited position ring plate 19. The inner circle edge of the left end of the main shell 5 is integrally formed with a concentric arrangement and sleeved on the right side of the second connecting sleeve 18. The limiting cylinder 20 on the side is rotatably connected to the second connecting sleeve 18, the right end of the limiting cylinder 20 extends into the interior of the main shell 5 and is pressed against the left annular surface of the limiting ring plate 19, the left end of the limiting cylinder 20 is located on the left side of the main shell 5, and the left end edge of the limiting cylinder 20 is provided with a plurality of slots 21 arranged in a circumferential array, and the outer circumference of the right end of the push rod is integrally formed with an annular boss 22, the inner cavity diameter of the second connecting sleeve 18 is larger than the outer diameter of the annular boss 22, and the inner diameter of the left port of the second connecting sleeve 18 is larger than the outer diameter of the push rod and smaller than the outer diameter of the annular boss 22. The metering ring 8 is a cylindrical structure that is transparent on the left and right, and the metering ring 8 is the same The central sliding sleeve is outside the first connecting sleeve 17, the inner diameter of the metering ring 8 is larger than the outer diameter of the limiting cylinder 20, the right end of the metering ring 8 is a trumpet mouth with a smaller left side and a larger right side, and the inner circumference of the right side of the metering ring 8 is integrally formed with a plurality of blocks 23 arranged in a circular array, and a limited sector plate 24 is integrally formed on the inner circumference of the metering ring 8 between two adjacent blocks 23, and a plurality of sliding through holes 25 arranged in a circular array and located between the first connecting sleeve 17 and the limiting cylinder 20 are opened on the outer circumference of the right side of the second connecting sleeve 18, and each limiting sector plate 24 corresponds to each sliding through hole 25 one by one, and each limiting sector plate 24 corresponds to each sliding through hole 25 one by one. The inner diameter of the plate 24 is smaller than the outer diameter of the first connecting sleeve 17. The inner edges of each limiting sector plate 24 are slidably arranged in each sliding through hole 25 respectively. Each clamping block 23 is clamped and matched with each clamping groove 21 respectively. An elastic clamping plate 26 is integrally formed in the middle of the right side edge of each sliding through hole 25. A triangular convex ridge 27 is arranged on the outside of the left side edge of the elastic clamping plate 26. The inner edges of each limiting sector plate 24 are arc protrusions. The left side edge of the elastic clamping plate 26 is inserted into the metering ring 8 and is located on the left side of the corresponding limiting sector plate 24. The triangular convex ridge 27 is clamped on the left side of the corresponding limiting sector plate 24.

[0078] The left end face of the gear box 12 is fixedly connected with a non-reversal stop 28 through a plurality of plastic columns 31. The right end of the second connecting sleeve 18 is sleeved on the left end of the pipe column 15 and rotatably connected together. The right end of the second connecting sleeve 18 is fixedly connected with a ratchet 29 sleeved on the outside of the pipe column 15. The non-reversal stop 28 is sleeved on the outside of the ratchet 29 and rotatably connected. The inner circle of the non-reversal stop 28 is provided with a ratchet groove 30 that matches and meshes with the outer circle ratchet teeth of the ratchet 29.

[0079] A sensing sensor 36 and a switch sensor 35 are provided on the left side of the inner wall of the main shell 5, and a sound sensor 33 and a vibration sensor 34 are provided inside the ratchet 29. The sensing sensor 36, the switch sensor 35, the sound sensor 33 and the vibration sensor 34 are all connected to the controller 6 through signal lines. The main shell 5 is provided with a charging interface (not shown in the figure) electrically connected to the micro battery life battery 7 and a display screen 32. The controller 6 has a built-in Bluetooth module, and the controller 6 is connected to the display screen 32 by signal.

[0080] The manual injection acoustic precision control metering method of the present invention comprises the following steps:

[0081] A. When the power of the injection pen is less than 20%, it will enter the energy-saving mode and can only be operated manually; when the user picks up the injection pen, the sensing sensor 36 arranged on the left side of the inner wall of the main shell 5 sends a signal to the controller 6, and the controller 6 receives the signal that someone wants to use the injection pen;

[0082] B. Remove the pen cap 1 and the needle cap 9, and then push the metering ring 8 to the left to separate the metering ring 8 from the left end of the main shell 5, so that each card block 23 is disengaged from each card slot 21, and the triangular convex ridge 27 on the left side of each elastic card plate 26 moves to the left to the right side of the corresponding limit sector plate 24. At this time, the main shell 5 rotates relative to the second connecting sleeve 18 and the pen core 2; the switch sensor 35 arranged on the left side of the inner wall of the main shell 5 detects that the metering ring 8 moves to the left, and transmits a signal to the controller 6. The controller 6 receives the signal and the manual injection mode is turned on;

[0083] C. The user inserts the needle 3 into the human body, holds the refill 2 and the metering ring 8 in one hand, and holds the main housing 5 in the other hand, and rotates the main housing 5, which drives the internal gear box 12 to rotate. The power output shaft of the gear box 12 is in a locked state at this time, and the power output shaft of the gear box 12 drives the screw 13 to rotate. The screw 13 drives the threaded hollow push rod 14 to move leftward, and the hollow push rod 14 pushes the injection piston 10 to move leftward in the refill 2. The injection piston 10 injects the insulin liquid to be injected in the refill 2 into the human body;

[0084] D. During the rotation of the main shell 5, the anti-reverse rotation 28 on the inner wall of the main shell 5 rotates relative to the ratchet wheel, and the ratchet groove 30 on the inner circle of the anti-reverse rotation 28 and the ratchet teeth on the outer circle of the ratchet wheel 29 will make a collision sound and vibrate at the same time. The sound sensor 33 and the vibration sensor 34 are arranged inside the ratchet wheel 29. Each collision sound signal is collected by the sound sensor 33, and each vibration signal is collected by the vibration sensor 34. The sound sensor 33 and the vibration sensor 34 transmit the collected signals to the controller 6. The controller 6 classifies and identifies the sound and vibration signals through an internal algorithm, records and stores the injection amount of insulin liquid. Each successful identification represents the injection of 0.5u of insulin liquid, thereby achieving accurate calculation of the injection amount during manual injection; the injection amount is displayed on the display screen 32, and the user only needs to observe the value on the display screen 32 to know the injection amount;

[0085] E. When the target injection volume is reached, stop rotating the main housing 5, and move the metering ring 8 to the right to reset it, so that each block 23 is correspondingly engaged in each slot 21, and the main housing 5 and the pen core 2 can be fixed; finally, close the needle cap 9 and the pen cap 1.

[0086] like Figure 4 and Figure 5 As shown, 9 ratchet grooves 30 are evenly arranged on the inner circle of the anti-reversal 28 along the circumferential direction, the central angle between two adjacent ratchet grooves 30 is 40°, the 9 ratchet grooves 30 are divided into three groups of the same order, namely group 1, group 2 and group 3, 3 adjacent ratchet grooves 30 form a group, the depth of the 3 ratchet grooves 30 in each group increases in sequence along the clockwise direction, the depth layout of the ratchet grooves 30 in the three groups is consistent, the ratchet teeth on the outer circle of the ratchet wheel 29 are evenly arranged along the circumferential direction, the central angle between two adjacent ratchet teeth is 120°, when the ratchet wheel 29 rotates, the three ratchet teeth collide with the ratchet groove 30 of the same depth at the same time, and the sound and vibration emitted when the ratchet groove 30 and the ratchet teeth collide are different when the ratchet groove 30 rotates 40° every time;

[0087] In step D, the internal algorithm of the controller 6 includes feature processing of the sound signal and the vibration signal when the ratchet groove 30 on the inner circle of the anti-reversal 28 collides with the ratchet teeth on the outer circle of the ratchet wheel 29. After the internal algorithm processing of the controller 6 and the neural network classification and recognition, the sound signal recognition result, the vibration signal recognition result and the detection time difference T between the sound signal and the vibration signal are uploaded to the support vector machine (SVM) for classification and recognition again.

[0088] The characteristic processing method of the sound signal when the ratchet groove 30 on the inner circle of the anti-reversal 28 collides with the ratchet teeth on the outer circle of the ratchet wheel 29 adopts the Mel reciprocal coefficient (MFCC) method. The MFCC is extracted based on the Mel frequency. The mapping relationship between the Mel frequency and the normal Hertz frequency adopts the formula (1):

[0089]

[0090] Where f Mel is the Mel frequency, the unit is MEL, f is the linear frequency, the unit is Hz;

[0091] like Figure 6 As shown, the specific process of the Mel Reciprocal Coefficient (MFCC) method includes the following steps:

[0092] 1) Collecting sound signals: When the ratchet groove 30 collides with the ratchet teeth on the outer circle of the ratchet wheel 29, the sound of the collision is transmitted to the controller 6 through the sound sensor 33 provided inside the ratchet wheel 29, wherein the sound signal collection time for each time is 0.5s;

[0093] 2) Sound preprocessing: The preprocessing process mainly involves pre-emphasis, framing and windowing of sound signals;

[0094] The pre-emphasis of the sound signal is usually performed by filtering. Here, the input signal x(n) is passed through a high-pass filter. The filter is V(n) = x(n) - a1x(n-1) in the time domain and H(z) = 1 - a1z in the frequency domain. -1 , where 0.9≤a1≤1.0, a1 is the pre-emphasis coefficient and is preferably 0.97;

[0095] When dividing the frame, the "short-term stable characteristics" of the sound signal are used to decompose the signal into several frames. Each frame can be set to 20ms, and the previous and next frames overlap by half. After dividing the frame, the time domain signal x(n) is combined into x i (n), i represents the relative frame number;

[0096] Windowing is to multiply the original function and the window function; let x(n) be the time domain signal and w(n) be the window function, then the N-point sequence x obtained by truncating x(n) with w(n) is n (n) See formula (2):

[0097] x n (n)=w(n)x(n) (2)

[0098] This patent selects the Hamming window, and formula (3) is as follows:

[0099]

[0100] Where W(n, a2) represents the window function, n=1,2,3,…,N is the window function sampling point number, a2 represents the middle position of the window, preferably 0.46;

[0101] 3) Fast Fourier Transform (FFT): for x i (n) After FFT, we get X i(K), by combining formula (2) with discrete Fourier transform (DFT) calculation formula (4), we can get X i (K) is calculated by formula (5), where the Hamming window is used:

[0102]

[0103] In formula (4), x(n) is the original signal, X N (k) is the coefficient after DCT transformation, and N is the number of points of the original signal;

[0104] Formula (5) N is the length of the DFT transform, k is the frequency, and h(n) is a window function with the same length as the sample;

[0105] 4) Mel filter bank: The Mel frequency filter bank is a group of triangular bandpass filters H arranged regularly within the specified frequency spectrum of the acoustic signal. m (k), 0≤m≤M, M is the number of filters, the number of filters is usually between 24 and 40, the center frequency of each triangular filter is f(m), and the transfer function formula (6) is as follows:

[0106]

[0107] Where H m (k) is the frequency response of the Mel filter, and the preferred number of M is 24;

[0108] 5) Obtaining logarithmic energy: In order to make the Mel spectrum obtained by the Mel filter bank more robust to noise and spectrum estimation errors, the logarithmic energy of the Mel spectrum is often obtained. The transfer function formula (7) from the linear spectrum to the logarithmic spectrum is as follows:

[0109]

[0110] Where X(k) is the linear spectrum, S(m) is the logarithmic spectrum, and m is the number of triangular bandpass filters;

[0111] 6) Discrete cosine transform: The logarithmic spectrum S(m) is subjected to discrete cosine transform (DCT) and can be converted back to the time domain. At this time, the time domain is the cepstrum domain, and the Mel frequency cepstrum coefficient can be obtained. The calculation formula (8) is as follows:

[0112]

[0113] Where C(b) represents the MFCC of the bth filter, L is the dimension of the extracted feature parameter MFCC; M is the number of Mel filter groups;

[0114] 7) Fully connected neural network (DNN) classification: The MFCC static characteristics of acoustic detection are input into the fully connected neural network (DNN) model, and the recognition results are output. The signals finally processed and recognized by the sound signals are recorded as voice 1, voice 2, voice 3 and voice 4 respectively; voice 1 is the sound coefficient generated by the collision between the ratchet groove 30 with the largest depth and the ratchet 29, voice 2 is the sound coefficient generated by the collision between the ratchet groove 30 with the second largest depth and the ratchet 29, voice 3 is the sound coefficient generated by the collision between the ratchet groove 30 with the smallest depth and the ratchet 29, and voice 4 is the sound coefficient generated by other external collisions. Sample training is performed based on the fully connected neural network. The sample factors include the depth of the ratchet groove 30, the service life of the insulin syringe, the season of insulin injection, and the ambient temperature of insulin injection. 2000 groups of samples are trained, and 500 groups of each voice 1, voice 2, voice 3 and voice 4 are trained. The fully connected neural network (DNN) model is a conventional model and will not be described in detail here.

[0115] The special diagnosis and processing method for the vibration signal when the ratchet groove 30 on the inner circle of the anti-reversal 28 collides with the ratchet teeth on the outer circle of the ratchet wheel 29 adopts the empirical mode decomposition method (EMD). EMD uses the time scale characteristics of the signal to decompose the signal without setting the basis function. It is essentially a time-frequency analysis method, which mainly completes the decomposition of a non-stationary signal. After decomposition, the linear sum of multiple IFMs and a trend term is obtained, that is, a non-stationary signal y(t) can be expressed as:

[0116]

[0117] Among them, IMF i (t) represents the i-th order intrinsic mode function, r(t) is the low-frequency pulsation, i.e., the trend term;

[0118] like Figure 7 As shown, the specific process of the empirical mode decomposition (EMD) method includes performing EMD and convolutional neural network (CNN) classification on a vibration signal in sequence.

[0119] The specific steps for performing EMD on a signal are:

[0120] ① Collecting vibration signals: When the ratchet groove 30 collides with the ratchet teeth on the outer circle of the ratchet wheel 29, the vibration signal of the collision is transmitted to the controller 6 through the vibration sensor 34 provided inside the ratchet wheel 29, wherein each vibration signal collection time is 0.5s;

[0121] ② Find all the maximum and minimum points of the vibration signal y(t), and then connect all the maximum and minimum points with a curve to obtain the upper envelope emax (t) and the lower envelope e min (t), and then find the mean envelope function d(t) of the upper envelope and the lower envelope:

[0122]

[0123] ③ In formula (10), the difference between y(t) and d(t) is h1(t). If h1(t) is not an IMF, continue the above process until the kth cycle h 1k (t) is an IMF, denoted as c1(t);

[0124] ④According to the decomposed IMF, the residual signal y1(t) is obtained:

[0125] y1(t)=y(t)-c1(t) (11)

[0126] ⑤ Repeat the above steps until the nth order h n (t) becomes a monotonic sequence, and the trend term r is defined n (t) is y n (t).

[0127] The specific process of convolutional neural network (CNN) classification is as follows: the vibration preprocessing signal of vibration detection is input into the convolutional neural network (CNN) model, and the recognition result is output; the vibration signal is finally processed and recognized as vibration 1, vibration 2, vibration 3 and vibration 4. Vibration 1 is the vibration processing signal generated by the collision between the ratchet groove 30 with the largest depth and the ratchet 29, vibration 2 is the vibration processing signal generated by the collision between the ratchet groove 30 with the second largest depth and the ratchet 29, vibration 3 is the vibration processing signal generated by the collision between the ratchet groove 30 with the largest depth and the ratchet 29, and vibration 4 is the vibration processing signal of other external collisions. Sample training is performed based on the fully connected neural network. The sample factors include the depth of the ratchet groove 30, the service life of the insulin syringe, the season of insulin injection and the ambient temperature of insulin injection. 2000 groups of samples are trained, and 500 groups of samples are trained for each type of vibration 1, vibration 2, vibration 3 and vibration 4. The convolutional neural network (CNN) model is a conventional model and will not be described in detail here.

[0128] The sound signal recognition result, the vibration signal recognition result and the detection time difference T between the sound signal and the vibration signal are uploaded to the support vector machine (SVM) for classification and recognition again as follows:

[0129] There are many kinds of sound signals and vibration signals in the process of manual insulin injection, so this is a multi-classification problem. In the process of sound detection and vibration detection, the fundamental frequency periods of sound signals and vibration signals are different, so we need to train SVM model samples according to actual conditions. Sample training is based on the SVM model. The sample factors include the depth of the spine groove 30, the service life of the insulin syringe, the season of insulin injection, the detection time difference T of the sound signal and the vibration signal, and the ambient temperature of insulin injection. 2000 groups of samples are trained;

[0130] like Figure 8 As shown, voice 1, voice 2, voice 3, voice 4 and vibration 1, vibration 2, vibration 3, vibration 4 are input into the SVM model, and the detection time difference T between the sound signal and the vibration signal is also input into the SVM model. The output results of the SVM model recognition are: normal injection information, other syringe injection information and invalid injection information. This can ensure the accuracy of manual insulin injection.

[0131] There are two types of manual injection metering cycles. The first type is the entire time period from the switch sensor 35 being turned on to being turned off. The second type is after the switch sensor 35 receives the signal of turning on, the manual injection device may not turn off the manual injection mode due to other circumstances. At this time, when the sensing sensor 36 does not sense any external contact for 10 minutes, the metering is automatically terminated. The amount of insulin injected each time is shown in Formula 12:

[0132] S=0.5b1+b2×0 (12)

[0133] Where S is the manual injection cycle, b1 is the number of injection device detection and identification times, and b2 is the number of invalid detection and identification times.

[0134] The first type of situation is as follows: when a user uses the insulin injection device for manual injection, he removes the pen cap 1 and the needle cap 9, and then pushes the metering ring 8 to the left to separate the metering ring 8 from the left end of the main shell 5. At this time, the switch sensor 35 transmits a signal to start manual injection, and the injection cycle begins. After the patient completes the manual injection, he pushes the metering ring 8 to the right to fit the metering ring 8 with the left end of the main shell 5. At this time, the switch sensor 35 transmits a signal to close the manual injection, and the injection cycle ends.

[0135] The second situation is as follows: when a patient uses an insulin injection device for manual injection, he removes the pen cap 1 and the needle cap 9, and then pushes the metering ring 8 to the left to separate the metering ring 8 from the left end of the main shell 5. At this time, the switch sensor 35 transmits a signal to start manual injection, and the injection cycle begins. When something unexpected happens during the patient's injection, the manual injection is not completed, and the manual injection mode is not turned off. At this time, the sensing sensor 36 senses that the patient has not touched the injection device within 10 minutes, and the metering cycle is automatically ended.

[0136] This embodiment does not impose any formal limitation on the shape, material, structure, etc. of the present invention. Any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention shall fall within the protection scope of the technical solution of the present invention.

Claims

1. A manually controlled injection pen, characterized in that: The invention comprises a pen cap, a pen core, a needle, a pen core sleeve, a main shell, a power assembly, a controller, a micro battery and a metering ring. The pen cap, the pen core, the needle, the pen core sleeve and the main shell are co-centered and are arranged horizontally in the left-right direction. The right end of the pen cap, the right end of the pen core, the right end of the pen core sleeve and the left end of the main shell are all open. The pen core sleeve is fixedly sleeved on the outside of the pen core, the pen cap is pluggable and sleeved on the outside of the pen core sleeve, the needle is fixedly installed at the center of the left end of the pen core sleeve, the right end of the needle passes through the pen core sleeve and the left end of the pen core and is connected to the inner The left end of the needle is located on the left side of the refill sleeve, the left end of the refill sleeve is plugged and connected with a needle cap sleeved on the outside of the needle, the right end of the refill is connected to the left end of the main shell through a connecting assembly, an injection piston is sealed and slidably arranged inside the refill, the power assembly, the controller and the micro battery life battery are all installed in the main shell, the metering ring is slidably sleeved on the outside of the connecting assembly, the power assembly drives the injection piston to move to the left, the micro battery life battery is electrically connected to the power assembly and the controller respectively, and the controller is connected to the power assembly signal; The power assembly includes an electric motor, a gear box, a screw and a hollow push rod. The electric motor is fixedly installed on the right side of the main shell, the gear box is fixedly installed in the middle of the main shell, the outer circumference of the gear box is clamped and fixed with the inner wall of the main shell, the screw is cocentrically rotatable and arranged on the left side of the main shell, the middle part of the left end face of the gear box is integrally formed with a pipe column, the right end of the screw cocentrically extends into the pipe column and is coaxially connected to the left end of the power output shaft in the center of the gear box as a whole, the left end of the screw cocentrically passes through the connecting assembly and is located on the right side of the right end of the pen core, the hollow push rod is cocentrically sleeved on the outside of the screw, the right end of the inner circle of the hollow push rod is provided with an internal thread connected with the external thread of the screw, the left end of the hollow push rod extends into the pen core and is fixedly connected with the right end face of the injection piston, the power shaft of the electric motor is horizontally arranged in the left and right directions, the left end of the power shaft of the electric motor is transmission-connected to the power input shaft of the gear box, the micro battery life battery is electrically connected to the electric motor, the controller is signal-connected to the electric motor, and a switch button for controlling the operation of the electric motor is provided in the middle part of the right side face of the main shell; The connecting assembly includes a first connecting sleeve and a second connecting sleeve. The first connecting sleeve is fixedly sleeved on the outside of the second connecting sleeve with the same center. The left end of the first connecting sleeve is pressed and contacted with the right end of the pen core. The right end of the pen core sleeve is sleeved on the left end of the first connecting sleeve and threaded together. The second connecting sleeve is sleeved on the outside of the hollow push rod with the same center. The right end of the second connecting sleeve extends into the left side of the main shell with the same center. The outer circle of the right end of the second connecting sleeve is integrally formed with a limited position ring plate. The inner circle edge of the left end of the main shell is integrally formed with a concentric setting and sleeved on the right side of the second connecting sleeve. The limiting cylinder on the side is rotatably connected with the second connecting sleeve, the right end of the limiting cylinder extends into the interior of the main shell and is pressed into contact with the left annular surface of the limiting ring plate, the left end of the limiting cylinder is located on the left side of the main shell, and the left end edge of the limiting cylinder is provided with a plurality of slots arranged in a circumferential array, and an annular boss is integrally formed on the outer circumference of the right end of the push rod, the inner cavity diameter of the second connecting sleeve is larger than the outer diameter of the annular boss, and the inner diameter of the left port of the second connecting sleeve is larger than the outer diameter of the push rod and smaller than the outer diameter of the annular boss. The metering ring is a cylindrical structure that is transparent on the left and right, and the metering ring is the same as the central The central sliding sleeve is arranged on the outside of the first connecting sleeve, the inner diameter of the metering ring is larger than the outer diameter of the limiting cylinder, the right end of the metering ring is a trumpet mouth with a smaller left side and a larger right side, the inner circumference of the right side of the metering ring is integrally formed with a plurality of blocks arranged in a circumferential array, and a limited position sector plate is integrally formed on the inner circumference of the metering ring between two adjacent blocks, and a plurality of sliding through holes arranged in a circumferential array and located between the first connecting sleeve and the limiting cylinder are opened on the outer circumference of the right side of the second connecting sleeve, and each limiting sector plate corresponds to each sliding through hole one by one. The inner diameter of the positioning sector plate is smaller than the outer diameter of the first connecting sleeve, the inner edges of each limiting sector plate are slidably arranged in each sliding through hole, each clamping block is clamped and matched with each clamping groove in a one-to-one manner, an elastic clamping plate is integrally formed in the middle of the right side of each sliding through hole, a triangular convex ridge is arranged on the outside of the left side of the elastic clamping plate, the inner edges of each limiting sector plate are arc convex, the left side of the elastic clamping plate is inserted into the metering ring and is located on the left side of the corresponding limiting sector plate, and the triangular convex ridge is clamped on the left side of the corresponding limiting sector plate; The left end face of the gear box is fixedly connected with a non-reversing device through a plurality of plastic columns, the right end of the second connecting sleeve is sleeved on the left end of the pipe column and rotatably connected together, the right end of the second connecting sleeve is fixedly connected with a ratchet sleeved on the outside of the pipe column, the non-reversing device is sleeved on the outside of the ratchet and rotatably connected, and the inner circle of the non-reversing device is provided with a ratchet groove that matches and meshes with the outer circle ratchet teeth of the ratchet; A sensing sensor and a switch sensor are arranged on the left side of the inner wall of the main shell, and a sound sensor and a vibration sensor are arranged inside the ratchet. The sensing sensor, the switch sensor, the sound sensor and the vibration sensor are all connected to the controller through a signal line. A charging interface and a display screen electrically connected to the micro battery life battery are arranged on the main shell. A Bluetooth module is built into the controller, and the controller is connected to the display screen by signal. The method for manual injection acoustic precision control metering using the injection pen comprises the following steps: A. When the power of the injection pen is less than 20%, it will enter the energy-saving mode and can only be operated manually. When the user picks up the injection pen, the sensing sensor set on the left side of the inner wall of the main shell sends a signal to the controller, and the controller receives the signal that someone wants to use the injection pen; B. Remove the pen cap and needle cap, and then push the metering ring to the left to separate the metering ring from the left end of the main shell, so that each card block is disengaged from each card slot, and the triangular ridges on the left side of each elastic card plate move to the left to the right side of the corresponding limit sector plate. At this time, the main shell rotates relative to the second connecting sleeve and the pen core; the switch sensor arranged on the left side of the inner wall of the main shell detects that the metering ring moves to the left, and transmits a signal to the controller. The controller receives the signal, and the manual injection mode is turned on; C. The user inserts the needle into the human body, holds the refill sleeve and metering ring with one hand, and holds the main housing with the other hand, and rotates the main housing, which drives the internal gear box to rotate. The power output shaft of the gear box is in a locked state at this time, and the power output shaft of the gear box drives the screw to rotate. The screw drives the threaded hollow push rod to move left, and the hollow push rod pushes the injection piston to move left in the refill. The injection piston injects the insulin liquid to be injected in the refill into the human body; D. During the rotation of the main housing, the anti-reverse rotation of the inner wall of the main housing rotates relative to the ratchet wheel, and the ratchet groove on the inner circle of the anti-reverse rotation and the ratchet teeth on the outer circle of the ratchet wheel will make a collision sound and vibrate at the same time. A sound sensor and a vibration sensor are arranged inside the ratchet wheel. Each collision sound signal is collected by the sound sensor, and each vibration signal is collected by the vibration sensor. The sound sensor and the vibration sensor transmit the collected signals to the controller. The controller classifies and identifies the sound and vibration signals through the internal algorithm, records and stores the injection amount of insulin liquid. Each successful identification represents the injection of 0.5ul of insulin liquid, so as to achieve accurate calculation of the injection amount during manual injection; the injection amount is displayed on the display screen, and the user only needs to observe the value on the display screen to know the injection amount; E. When the target injection volume is reached, stop rotating the main housing and move the metering ring to the right to reset it, so that each card block is correspondingly connected to each card slot, and the main housing and the pen core can be fixed; finally, close the needle cap and the pen cap; The inner circle of the anti-reversal wheel is evenly provided with 9 ratchet grooves along the circumferential direction, and the central angle between two adjacent ratchet grooves is 40°. The 9 ratchet grooves are divided into three groups of the same order, namely group 1, group 2 and group 3. Three adjacent ratchet grooves form a group. The depths of the three ratchet grooves in each group are different, and they increase in sequence in a clockwise direction. The depth layout of the three groups of ratchet grooves is consistent. The outer circle of the ratchet wheel has 3 ratchet teeth evenly provided along the circumferential direction, and the central angle between two adjacent ratchet teeth is 120°. When the ratchet wheel rotates, the three ratchet teeth collide with the ratchet grooves of the same depth at the same time. When the ratchet groove rotates 40°, the sound and vibration emitted when the ratchet grooves collide with the ratchet teeth are different. The internal algorithm of the controller in step D includes feature processing of the sound signal and vibration signal when the ratchet groove on the inner circle of the anti-reversal wheel collides with the ratchet teeth on the outer circle of the ratchet wheel. After the internal algorithm processing of the controller and the neural network classification and recognition, the sound signal recognition result, the vibration signal recognition result and the detection time difference T between the sound signal and the vibration signal are uploaded to the support vector machine SVM for classification and recognition again.

2. The manually controlled injection pen according to claim 1, characterized in that: The characteristic processing method of the sound signal when the ratchet groove on the inner circle of the anti-reversal wheel collides with the ratchet teeth on the outer circle of the ratchet wheel adopts the Mel reciprocal coefficient MFCC method. MFCC is extracted based on the Mel frequency. The mapping relationship between the Mel frequency and the normal Hertz frequency adopts formula (1): Where f Mel is the Mel frequency, the unit is MEL, f is the linear frequency, the unit is Hz; The specific process of the Mel reciprocal coefficient MFCC method includes the following steps: 1) Collecting sound signals: When the ratchet groove collides with the ratchet teeth on the outer circle of the ratchet wheel, the sound of the collision is transmitted to the controller through the sound sensor set inside the ratchet wheel. The sound signal collection time for each time is 0.2-0.5s; 2) Sound preprocessing: The preprocessing process mainly involves pre-emphasis, framing and windowing of sound signals; The pre-emphasis of the sound signal uses filtering. Here, the input signal x(n) is passed through a high-pass filter. The filter is V(n) = x(n)-a1x(n-1) in the time domain and H(z) = 1-a1z in the frequency domain. -1 , a1 is the pre-emphasis coefficient and is 0.97; When dividing the frame, the short-term stable characteristics of the sound signal are used to decompose the signal into several frames. Each frame is set to 20ms, and the previous and next frames overlap by half. After dividing the frame, the time domain signal x(n) is combined into x i (n), i represents the relative frame number; Windowing is to multiply the original function and the window function; suppose x(n) is the time domain signal and w(n) is the window function. Then the N-point sequence x(n) obtained by truncating x(n) with w(n) is shown in formula (2): x n (n)=w(n)x(n) (2) Here, the Hamming window is used, and formula (3) is as follows: Where W(n, a2) represents the window function, n=1,2,3,…,N is the window function sampling point number, a2 represents the middle position of the window and is 0.46; 3) Fast Fourier Transform FFT: for x i (n) After FFT, we get X i (K), by combining formula (2) with discrete Fourier transform DFT calculation formula (4), we can get X i (K) is calculated by formula (5), where the Hamming window is used: In formula (4), x(n) is the original signal, X N (k) is the coefficient after DCT transformation, and N is the number of points of the original signal; Formula (5) N is the length of the DFT transform, k is the frequency, and h(n) is a window function with the same length as the sample; 4) Mel filter bank: The Mel frequency filter bank is a group of triangular bandpass filters H arranged regularly within the specified frequency spectrum of the acoustic signal. m (k), 0≤m≤M, M is the number of filters, the center frequency of each triangular filter is f(m), and the transfer function formula (6) is as follows: Where H m (k) is the frequency response of the Mel filter, and the number of M is 24; 5) Obtaining logarithmic energy: In order to make the Mel spectrum obtained by the Mel filter bank more robust to noise and spectrum estimation errors, the logarithmic energy of the Mel spectrum is obtained. The transfer function formula (7) from the linear spectrum to the logarithmic spectrum is as follows: Where X(k) is the linear spectrum, S(m) is the logarithmic spectrum, and m is the number of triangular bandpass filters; 6) Discrete cosine transform: The logarithmic spectrum S(m) is subjected to discrete cosine transform DCT, which can be converted back to the time domain. At this time, the time domain is the cepstrum domain, and the Mel frequency cepstrum coefficient can be obtained. The calculation formula (8) is as follows: Where C(b) represents the MFCC of the bth filter, L is the dimension of the extracted feature parameter MFCC; M is the number of Mel filter groups; 7) Fully connected neural network DNN classification: The MFCC static characteristics of acoustic detection are input into the fully connected neural network DNN model, and the recognition results are output. The signals finally processed and recognized by the sound signals are recorded as voice 1, voice 2, voice 3 and voice 4 respectively; voice 1 is the sound coefficient generated by the collision between the ratchet groove with the largest depth and the ratchet, voice 2 is the sound coefficient generated by the collision between the ratchet groove with the second largest depth and the ratchet, voice 3 is the sound coefficient generated by the collision between the ratchet groove with the smallest depth and the ratchet, and voice 4 is the sound coefficient generated by other external collisions; sample training is performed based on the fully connected neural network. The sample factors include the depth of the ratchet groove, the service life of the insulin syringe, the season of insulin injection and the ambient temperature of insulin injection. 2000 groups of samples are trained, and 500 groups of each type of voice 1, voice 2, voice 3 and voice 4 are trained.

3. The manually controlled injection pen according to claim 1, characterized in that: The special diagnosis and processing method for the vibration signal when the ratchet groove on the inner circle of the anti-reversal wheel collides with the ratchet teeth on the outer circle of the ratchet wheel adopts the empirical mode decomposition method EMD. EMD uses the time scale characteristics of the signal to decompose the signal without setting the basis function. It is essentially a time-frequency analysis method, which mainly completes the decomposition of a non-stationary signal. After decomposition, the linear sum of multiple IFMs and a trend term is obtained, that is, a non-stationary signal y(t) is expressed as: Among them, IMF i (t) represents the i-th order intrinsic mode function, r(t) is the low-frequency pulsation, i.e., the trend term; The specific process of the empirical mode decomposition method EMD includes performing EMD and convolutional neural network (CNN) classification on a vibration signal in sequence.

4. The manually controlled injection pen according to claim 3, characterized in that: The specific steps for performing EMD on a signal are: ① Collect vibration signals: When the ratchet groove collides with the ratchet teeth on the outer circle of the ratchet wheel, the vibration signal of the collision is transmitted to the controller through the vibration sensor set inside the ratchet wheel. The collection time of each vibration signal is 0.2-0.5s; ② Find all the maximum and minimum points of the vibration signal y(t), and then connect all the maximum and minimum points with a curve to obtain the upper envelope e max (t) and the lower envelope e min (t), and then find the mean envelope function d(t) of the upper envelope and the lower envelope: ③ In formula (10), the difference between y(t) and d(t) is h1(t). If h1(t) is not an IMF, continue the above process until the kth cycle h 1k (t) is an IMF, denoted as c1(t); ④According to the decomposed IMF, the residual signal y1(t) is obtained: y1(t)=y(t)-c1(t) (11) ⑤ Repeat the above steps until the nth order h n (t) becomes a monotonic sequence, and the trend term r is defined n (t) is y n (t).

5. The manually controlled injection pen according to claim 4, characterized in that: The specific process of convolutional neural network (CNN) classification is as follows: input the vibration preprocessing signal of vibration detection into the convolutional neural network (CNN) model and output the recognition result; the vibration signal is finally processed and recognized as vibration 1, vibration 2, vibration 3, vibration 4, Vibration 1 is the vibration processing signal generated by the collision between the ratchet groove with the largest depth and the ratchet, vibration 2 is the vibration processing signal generated by the collision between the ratchet groove with the second largest depth and the ratchet, vibration 3 is the vibration processing signal generated by the collision between the ratchet groove with the largest depth and the ratchet, and vibration 4 is the vibration processing signal of other external collisions. Sample training is performed based on a fully connected neural network. The factors of the samples include the depth of the ratchet groove, the service life of the insulin syringe, the season for insulin injection, and the ambient temperature for insulin injection. 2000 groups of samples are trained, and 500 groups of samples are trained for each type of vibration 1, vibration 2, vibration 3, and vibration 4.

6. The manually controlled injection pen according to claim 5, characterized in that: The sound signal recognition result, the vibration signal recognition result and the detection time difference T between the sound signal and the vibration signal are uploaded to the support vector machine SVM for classification and recognition again as follows: In the process of sound detection and vibration detection, the fundamental frequency periods of the sound signal and the vibration signal are different, so we need to train the SVM model samples according to the actual situation. The sample training is based on the SVM model. The sample factors include the depth of the spine groove, the service life of the insulin syringe, the season of insulin injection, the detection time difference T between the sound signal and the vibration signal, and the ambient temperature of insulin injection. 2000 groups of samples are trained; The voice 1, voice 2, voice 3, voice 4 finally processed and identified by the sound signal and the vibration 1, vibration 2, vibration 3, vibration 4 finally processed and identified by the vibration signal are input into the SVM model, and then the detection time difference T between the sound signal and the vibration signal is also input into the SVM model. The output results are: normal injection information, other syringe injection information and invalid injection information through SVM model recognition; this can ensure the accuracy of manual injection of insulin injection; There are two types of manual injection metering cycles. The first type is the entire time period from the switch sensor being turned on to being turned off. The second type is after the switch sensor is turned on, the manual injection device does not turn off the manual injection mode due to other reasons. At this time, when the sensor does not sense external contact for 10 minutes, the metering is automatically terminated. The amount of insulin injected each time is shown in formula (12): S=0.5b1+b2×0 (12) Where S is the injection volume of each injection cycle, b1 is the number of detection and identification times of the injection device, and b2 is the number of invalid detection and identification times.

7. The manually controlled injection pen according to claim 6, characterized in that: The first situation is as follows: when a user uses an insulin injection device for manual injection, he removes the pen cap and needle cap, and then pushes the metering ring to the left to separate the metering ring from the left end of the main shell. At this time, the switch sensor transmits a signal to start manual injection, and the injection cycle begins. After the patient completes the manual injection, he pushes the metering ring to the right to fit the metering ring with the left end of the main shell. At this time, the switch sensor transmits a signal to close manual injection, and the injection cycle ends. The second type of situation is as follows: when a patient is using an insulin injection device for manual injection, he removes the pen cap and needle cap, and then pushes the metering ring to the left to separate the metering ring from the left end of the main shell. At this time, the switch sensor transmits a signal to start manual injection, and the injection cycle begins. When something unexpected happens to the patient in the middle of the injection, he does not complete the manual injection, nor does he turn off the manual injection mode. At this time, the sensing sensor senses that the patient has not touched the injection device within 10 minutes, and the metering cycle automatically ends.

Images