Power device residual life prediction method based on node temperature difference and neural network

Abstract

The invention discloses a power device residual life prediction method based on a node temperature difference and a neural network, and the method specifically comprises the steps: collecting voltage and current signals of a traction converter according to a power loss model of a power device, and calculating the power loss of the power device; further, the power loss is used as an input signal, and junction temperature information of the power device is obtained by using a Foster model; then, according to the obtained junction temperature information of the power device, the temperature difference of two nodes is extracted; on the basis, node temperature difference data of the power device in different aging states are obtained through a power cycle experiment, and a cumulative damage calculation model is established according to the node temperature difference data; on the basis, effective prediction of the remaining service life of the power device is realized by using a BP neural network. The method is easy to implement, small in calculation burden, free of adverse influence of device aging and high in prediction precision, and therefore the method is suitable for prediction of the remaining service life of the power device under the long time scale, and the problems existing in the prior art are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of reliability assessment of power devices, and in particular relates to a method for predicting the remaining life of a power device based on node temperature difference and a neural network. Background Art

[0002] As the core equipment for energy conversion in trains, the reliable operation of traction converters significantly determines the running quality of trains. However, since traction converters frequently operate under high-power traction and braking conditions, power devices experience significant temperature rise during operation. In addition, variable operating conditions cause power devices to frequently experience severe electrical and thermal shocks, accelerating the fatigue failure process of power devices and seriously threatening the safe and reliable operation of traction converters.

[0003] At present, the maintenance method of trains is mostly "mileage maintenance", that is, the trains are inspected according to the mileage or service life of the trains (for example, the maintenance mileage of Fuxing trains is about 1.65 million kilometers). However, this maintenance method inspects all trains at the same time, and the operation and maintenance efficiency needs to be further improved. In addition, this maintenance mode may cause waste of manpower and material resources, and the operation and maintenance costs are high, which makes it difficult to meet the reliability requirements of traction inverters. Therefore, it is of great significance to implement accurate prediction of the remaining service life of power devices and carry out targeted maintenance work based on this, in order to ensure the safe and reliable operation of traction inverters.

[0004] Among the many remaining service life prediction methods for power devices, the remaining service life prediction method based on mission profile is highly favored due to its simple principle and easy implementation. However, the accuracy of this method depends on the accurate calculation of the mission profile and has high requirements for the life prediction model. In addition, this method is difficult to effectively characterize the aging process of power devices, resulting in the life prediction performance being easily affected by device aging. Summary of the invention

[0005] In view of the shortcomings of the existing remaining service life prediction method based on mission profile, the present invention provides a remaining service life prediction method for power devices based on node temperature difference and neural network.

[0006] A method for predicting the remaining life of a power device based on node temperature difference and a neural network of the present invention comprises the following steps:

[0007] Step 1: Calculate the node temperature difference of the power device based on the loss characteristics of the traction converter.

[0008] Power device losses are divided into conduction losses and switching losses.

[0009] In the traction converter, the conduction loss of the power device is expressed as:

[0010]

[0011] Where: P con 、V ce ,I c , M and They are conduction loss, collector-emitter voltage, collector current, modulation ratio and power factor angle respectively.

[0012] The switching loss of the power device is calculated as:

[0013]

[0014] Where: f sw 、E on+off ,I ref , U dc , U ref 、T j 、T ref , K V and C T They are switching frequency, switching energy, current reference value, DC side voltage, DC side voltage reference value, power device temperature, power device reference temperature, voltage correction factor and temperature correction factor.

[0015] Combining equations (1) and (2), the total power device loss is expressed as:

[0016]

[0017] Where: n and ω are the number of iteration cycles and the fundamental angular frequency, and:

[0018] ω=2πf (4)

[0019] Where: f is the fundamental frequency.

[0020] According to formula (3), the maximum value of power device loss is defined as:

[0021] P M =π(aV ce +b) (5)

[0022] According to the electrothermal analogy theory, in the first half of the first half-sine cycle, the temperature response of the power device is expressed as:

[0023]

[0024] Where: ΔT(t) is the temperature difference between the node temperature and the ambient temperature.

[0025] From formula (6), we get:

[0026]

[0027] Where: Rth and τ th are thermal resistance and temperature response time constant respectively, and:

[0028] τ th =R th C th (8)

[0029] At the node t = 1 / (2f), the temperature difference is calculated as:

[0030]

[0031] In the second half of a fundamental cycle, the power device only dissipates heat; therefore, the temperature difference at the end node of the first iteration cycle, i.e., t = 2 / 2f, is:

[0032]

[0033] Furthermore, when the temperature difference of the power device has not reached a steady state, the temperature of the power device will fluctuate and change with the iterative loss cycle. For the convenience of analysis, the device power loss curve is divided into m parts. For time t m The temperature difference ΔT m Calculated as:

[0034]

[0035] Where: ΔT m-1 , P m and P m-1 The time t m-1 The temperature difference, time t m The power loss and time t m-1 of power loss.

[0036] The half-cycle node temperature difference and full-cycle node temperature difference in each cycle are calculated as:

[0037]

[0038] As the iteration cycle increases, the temperature difference at node t = (2n-1) / (2f) and node t = (2n) / (2f) is calculated as:

[0039]

[0040] When it tends to steady state, n→∞, the temperature difference between node t=(2n-1) / (2f) and node t=(2n) / (2f) is expressed as:

[0041]

[0042] Where: ΔT half-cyc and ΔT end-cycThey are the half-cycle node temperature difference and the full-cycle node temperature difference in steady state respectively.

[0043] Substituting formula (9) into formula (14), we get:

[0044]

[0045] Step 2: Based on the node temperature difference obtained in step 1, analyze the characteristics of the node temperature difference under different aging modes.

[0046] The aging of the solder layer causes the heat dissipation path to shrink, which in turn increases the thermal resistance of the power device. After considering the aging of the solder layer, the thermal resistance of the power device is expressed as:

[0047] R thag =kR th (16)

[0048] Where: R thag and k are the thermal resistance and gain of the power device considering the aging of the solder layer.

[0049] Substituting formula (16) into formula (15), we get:

[0050]

[0051] Where: ΔT half-cyc-ags is the node temperature difference considering the aging of the solder layer.

[0052] Combining equation (15) and equation (17), we have:

[0053]

[0054] Where: half-cyc-ags Aging characteristic factor.

[0055] Similarly:

[0056]

[0057] Where: end-cyc-ags Aging characteristic factor.

[0058] When the bonding wire of the power device ages, its bonding resistance will increase significantly, which will cause the collector-emitter voltage V ce Increase; after considering the aging of the bonding wire, the collector-emitter voltage of the power device is expressed as:

[0059] V ceag =μV ce (20)

[0060] Where: V ceag and μ are the collector-emitter voltage and gain of the power device taking into account the aging of the bonding wire.

[0061] Substituting formula (20) into formula (5), we have:

[0062] P Mag =π(aV ceag +b)=π(aμV ce +b) (21)

[0063] Furthermore, the node temperature difference considering the aging of the bonding wire is calculated as:

[0064]

[0065] From formula (22), we get:

[0066]

[0067] Similarly:

[0068]

[0069] According to equations (23) and (24), when the bonding wire ages, ΔT half-cyc and ΔT end-cyc will increase, and the increase amplitude is the same.

[0070] In summary, ΔT half-cyc and ΔT end-cyc It can effectively characterize the fatigue failure of power devices and is therefore used as a performance degradation parameter to predict the remaining service life.

[0071] Step 3: Use power cycle experiments to obtain node temperature difference data of power devices under different aging conditions, and use this to establish a cumulative damage calculation model; on this basis, use BP neural network to effectively predict the remaining service life of power devices.

[0072] The beneficial technical effects of the present invention are:

[0073] 1. The present invention uses node temperature difference as a performance degradation parameter for remaining service life prediction, which can effectively reduce the adverse effects of device aging on remaining service life prediction.

[0074] 2. The present invention adopts a nonlinear model as a power device cumulative damage calculation model, which can accurately describe the fatigue failure process of the power device.

[0075] 3. The present invention adopts BP neural network to predict the remaining service life, which significantly improves the prediction accuracy and avoids the problem of fitting multiple model parameters.

[0076] Fourth, the remaining service life prediction method proposed in the present invention has good adaptability to power devices with different aging degrees, and has high prediction accuracy, which can meet the reliability requirements of traction converters. In addition, the method has good versatility and can also be transplanted to the remaining service life prediction method of other wide bandgap devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 The present invention is a flow chart of a method for predicting the remaining life of a power device based on node temperature difference and neural network.

[0078] Figure 2 It is a schematic diagram of the conversion of equivalent power loss to device temperature in the present invention.

[0079] Figure 3 It is a schematic diagram of power loss and node temperature difference in the present invention.

[0080] Figure 4 It is a variation trend diagram of the aging characteristic factor at different fundamental frequencies after considering the influence of solder layer aging in the present invention.

[0081] Figure 5 This is a variation trend diagram of the aging characteristic factor at different fundamental frequencies after considering the influence of bonding wire aging in the present invention.

[0082] Figure 6 It is a structural diagram of the BP neural network implemented by the present invention. DETAILED DESCRIPTION

[0083] The present invention is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0084] The process of the method for predicting the remaining life of a power device based on node temperature difference and neural network of the present invention is as follows: Figure 1 As shown in the figure, according to the power loss model of power devices, the voltage and current signals of the traction converter are collected to calculate the power loss of the power devices; further, the power loss is used as the input signal, and the junction temperature information of the power devices is obtained using the Foster model; then, based on the obtained junction temperature information of the power devices, the temperature difference between the two nodes is extracted; on this basis, the node temperature difference data of the power devices under different aging states is obtained by power cycle experiments, and the cumulative damage calculation model is established based on this. On this basis, the BP neural network is used to effectively predict the remaining service life of the power devices.

[0085] The method for predicting the remaining service life of a power device based on node temperature difference and neural network of the present invention uses node temperature difference and BP neural network to complete the prediction of the remaining service life of a power device at different aging degrees. The specific steps are:

[0086] Step 1: Calculate the node temperature difference of the power device based on the loss characteristics of the traction converter.

[0087] Power device losses are divided into conduction losses and switching losses.

[0088] In the traction converter, the conduction loss of the power device is expressed as:

[0089]

[0090] Where: P con 、V ce ,I c , M and They are conduction loss, collector-emitter voltage, collector current, modulation ratio and power factor angle respectively.

[0091] The switching loss of the power device is calculated as:

[0092]

[0093] Where: f sw 、E on+off ,I ref , U dc , U ref 、T j 、T ref , K V and C T They are switching frequency, switching energy, current reference value, DC side voltage, DC side voltage reference value, power device temperature, power device reference temperature, voltage correction factor and temperature correction factor.

[0094] Combining equations (1) and (2), the total power device loss is expressed as:

[0095]

[0096] Where: n and ω are the number of iteration cycles and the fundamental angular frequency, and:

[0097] ω=2πf (4)

[0098] Where: f is the fundamental frequency.

[0099] According to formula (3), the maximum value of power device loss is defined as:

[0100] P M =π(aV ce +b) (5)

[0101] According to the electrothermal analogy theory, in the first half of the first half-sine cycle, the temperature response of the power device is expressed as:

[0102]

[0103] Where: ΔT(t) is the temperature difference between the node temperature and the ambient temperature.

[0104] From formula (6), we get:

[0105]

[0106] Where: R th and τ th are thermal resistance and temperature response time constant respectively, and:

[0107] τ th =R th C th (8)

[0108] At the node t = 1 / (2f), the temperature difference is calculated as:

[0109]

[0110] In the second half of a fundamental cycle, the power device only dissipates heat; therefore, the temperature difference at the end node of the first iteration cycle, i.e., t = 2 / 2f, is:

[0111]

[0112] Furthermore, when the temperature difference of the power device has not reached a steady state, the temperature of the power device will fluctuate and change iteratively with the loss cycle; for the convenience of analysis, Figure 2 As shown, the device power loss curve is divided into m equal parts; for time t m The temperature difference ΔT m Calculated as:

[0113]

[0114] Where: ΔT m-1 , P m and P m-1 The time t m-1 The temperature difference, time t m The power loss and time t m-1 of power loss.

[0115] like Figure 3 As shown, the half-cycle node temperature difference and the full-cycle node temperature difference in each cycle are calculated as:

[0116]

[0117] like Figure 3As shown, as the iteration period increases, the temperature difference at node t = (2n-1) / (2f) and node t = (2n) / (2f) is calculated as:

[0118]

[0119] like Figure 3 As shown, when it tends to steady state, at this time n→∞, the temperature difference between node t=(2n-1) / (2f) and node t=(2n) / (2f) is expressed as:

[0120]

[0121] Where: ΔT half-cyc and ΔT end-cyc They are the half-cycle node temperature difference and the full-cycle node temperature difference in steady state respectively.

[0122] Substituting formula (9) into formula (14), we get:

[0123]

[0124] Step 2: Based on the node temperature difference obtained in step 1, analyze the characteristics of the node temperature difference under different aging modes.

[0125] The aging of the solder layer causes the heat dissipation path to shrink, which in turn increases the thermal resistance of the power device. After considering the aging of the solder layer, the thermal resistance of the power device is expressed as:

[0126] R thag =kR th (16)

[0127] Where: R thag and k are the thermal resistance and gain of the power device considering the aging of the solder layer.

[0128] Substituting formula (16) into formula (15), we get:

[0129]

[0130] Where: ΔT half-cyc-ags is the node temperature difference considering the aging of the solder layer.

[0131] Combining equation (15) and equation (17), we have:

[0132]

[0133] Where: half-cyc-ags is the aging characteristic factor, and its changing trend is as follows Figure 4 shown.

[0134] Similarly:

[0135]

[0136] Where: end-cyc-ags is the aging characteristic factor, and its changing trend is as follows Figure 4 shown.

[0137] When the bonding wire of the power device ages, its bonding resistance will increase significantly, which will cause the collector-emitter voltage V ce Increase; after considering the aging of the bonding wire, the collector-emitter voltage of the power device is expressed as:

[0138] V ceag =μV ce (20)

[0139] Where: V ceag and μ are the collector-emitter voltage and gain of the power device taking into account the aging of the bonding wire.

[0140] Substituting formula (20) into formula (5), we have:

[0141] P Mag =π(aV ceag +b)=π(aμV ce +b) (21)

[0142] Furthermore, the node temperature difference considering the aging of the bonding wire is calculated as:

[0143]

[0144] From formula (22), we get:

[0145]

[0146] Similarly:

[0147]

[0148] The changing trend of aging characteristic factors is as follows Figure 5 shown.

[0149] According to equations (23) and (24), when the bonding wire ages, ΔT half-cyc and ΔT end-cyc will increase, and the increase amplitude is the same.

[0150] In summary, ΔT half-cyc and ΔT end-cyc It can effectively characterize the fatigue failure of power devices and is therefore used as a performance degradation parameter to predict the remaining service life.

[0151] Step 3: Use power cycle experiments to obtain node temperature difference data of power devices under different aging conditions, and use this to establish a cumulative damage calculation model; on this basis, use a BP neural network (the BP neural network structure of the present invention is as follows Figure 6 (as shown) to achieve effective prediction of the remaining service life of power devices.

[0152] The method of the present invention is easy to implement, has a small computational burden, is not adversely affected by device aging, and has high prediction accuracy. Therefore, it is suitable for predicting the remaining service life of power devices on a long time scale, and solves the technical problem that the existing remaining service life prediction method cannot be adaptively adjusted according to the degree of device aging, resulting in the prediction error gradually increasing with the deepening of aging.

[0153] The method for predicting the remaining service life of a power device provided by the present invention fully considers the influence of device aging on the remaining service life prediction, and adopts the node temperature difference as the performance degradation parameter to reduce the adverse effects of device aging; in view of the problems of low generalization and prediction accuracy of the existing remaining service life prediction model, a BP neural network is adopted to realize the remaining service life prediction of the power device, which significantly improves the prediction accuracy and avoids the problem of fitting multiple model parameters.

[0154] The present invention can realize the prediction of the remaining service life of power devices on a long time scale. More importantly, compared with the remaining service life prediction method based on the analytical model, the method studied is not adversely affected by device aging, and has a higher prediction accuracy, which solves the technical problem that the existing remaining service life prediction method cannot be adaptively adjusted according to device aging, resulting in the prediction error gradually increasing with the deepening of aging. The remaining service life prediction method adopted can also be further extended and applied to wide bandgap devices (such as SiC devices and GaN devices).

Claims

1. A method for predicting the remaining life of a power device based on node temperature difference and neural network, characterized in that: The following steps are involved: Step 1: Calculate the node temperature difference of the power device according to the loss characteristics of the traction converter; Power device losses are divided into conduction losses and switching losses; In the traction converter, the conduction loss of the power device is expressed as: Where: P con 、V ce ,I c , M and They are conduction loss, collector-emitter voltage, collector current, modulation ratio and power factor angle respectively; The switching loss of the power device is calculated as: Where: f sw 、E on+off ,I ref , U dc , U ref , T j , T ref , K V and C T They are switching frequency, switching energy, current reference value, DC side voltage, DC side voltage reference value, power device temperature, power device reference temperature, voltage correction factor and temperature correction factor; Combining equations (1) and (2), the total power device loss is expressed as: Where: n and ω are the number of iteration cycles and the fundamental angular frequency, and: ω=2πf (4) Where: f is the fundamental frequency; According to formula (3), the maximum value of power device loss is defined as: P M =π(aV ce +b) (5) According to the electrothermal analogy theory, in the first half of the first half-sine cycle, the temperature response of the power device is expressed as: Where: ΔT(t) is the temperature difference between the node temperature and the ambient temperature; From formula (6), we get: Where: R th and τ th are thermal resistance and temperature response time constant respectively, and: τ th =R th C th (8) At the node t = 1 / (2f), the temperature difference is calculated as: In the second half of a fundamental cycle, the power device only dissipates heat; therefore, the temperature difference at the end node of the first iteration cycle, i.e., t = 2 / 2f, is: Furthermore, when the temperature difference of the power device has not reached a steady state, the temperature of the power device will fluctuate and change with the iterative loss cycle. For the convenience of analysis, the device power loss curve is divided into m parts. For time t m The temperature difference ΔT m Calculated as: Where: ΔT m-1 , P m and P m-1 The time t m-1 The temperature difference, time t m The power loss and time t m-1 Power loss; The half-cycle node temperature difference and full-cycle node temperature difference in each cycle are calculated as: As the iteration period increases, the temperature difference at node t = (2n-1) / (2f) and node t = (2n) / (2f) is calculated as: When it tends to steady state, n→∞, the temperature difference between node t=(2n-1) / (2f) and node t=(2n) / (2f) is expressed as: Where: ΔT half-cyc and ΔT end-cyc They are the half-cycle node temperature difference and the full-cycle node temperature difference in steady state respectively; Substituting formula (9) into formula (14), we get: Step 2: According to the node temperature difference obtained in step 1, the characteristics of the node temperature difference under different aging modes are analyzed; The aging of the solder layer causes the heat dissipation path to shrink, which in turn increases the thermal resistance of the power device. After considering the aging of the solder layer, the thermal resistance of the power device is expressed as: R thag =kR th (16) Where: R thag and k are the thermal resistance and gain of the power device considering the aging of the solder layer, respectively; Substituting formula (16) into formula (15), we get: Where: ΔT half-cyc-ags To consider the node temperature difference due to solder layer aging; Combining equation (15) and equation (17), we have: Where: half-cyc-ags is the aging characteristic factor; Similarly: Where: end-cyc-ags is the aging characteristic factor; When the bonding wire of the power device ages, its bonding resistance will increase significantly, which will cause the collector-emitter voltage V ce Increase; after considering the aging of the bonding wire, the collector-emitter voltage of the power device is expressed as: V ceag =μV ce (20) Where: V ceag and μ are the collector-emitter voltage and gain of the power device considering the aging of the bonding wire; Substituting formula (20) into formula (5), we have: P Mag =π(aV ceag +b)=π(aμV ce +b) (21) Furthermore, the node temperature difference considering the aging of the bonding wire is calculated as: From formula (22), we get: Similarly: According to equations (23) and (24), when the bonding wire ages, ΔT half-cyc and ΔT end-cyc will all increase, and the increase amplitude is the same; In summary, ΔT half-cyc and ΔT end-cyc It can effectively characterize the fatigue failure of power devices and thus be used as a performance degradation parameter for the prediction of remaining service life; Step 3: Use power cycle experiments to obtain node temperature difference data of power devices under different aging conditions, and use this to establish a cumulative damage calculation model; on this basis, use BP neural network to effectively predict the remaining service life of power devices.

Images