Diagnostic system and diagnostic method for determining state of compressed gas tank made of fiber-reinforced plastic

Abstract

The invention relates to a diagnostic system (103) for determining the state of a compressed gas tank (101) made of fiber-reinforced plastic. The diagnostic system (103) comprises an excitation element (105), a sensor (109) and an analysis unit (113), the excitation element (109) is configured to introduce sound waves into the compressed gas tank (101), the sensor (109) is configured to detect sound waves introduced by the excitation element (105) into the compressed gas tank (101), and the analysis unit (113) is configured to analyze the sound waves introduced by the excitation element (105) into the compressed gas tank (101). According to the invention, a characteristic value is assigned to the respective measured value determined by the sensor (109), said characteristic value indicating the state of the compressed gas tank (101) and the assigned state is output to an output unit.

Description

Technical Field

[0001] The invention relates to a diagnostic system and a diagnostic method for ascertaining the state of a compressed gas tank made of fiber-reinforced plastic, as well as a tank system having the proposed diagnostic system. Background Art

[0002] Compressed gas tanks for storing fluids such as hydrogen, for example, are usually made of fiber-reinforced plastics, that is to say of a matrix made of, for example, plastic and fibers embedded therein.

[0003] Studies have shown that defective compressed gas tanks are usually caused by two damage mechanisms. On the one hand, there are fiber breaks, in which the fibers are damaged, for example due to mechanical loads, so that the compressed gas tank usually shows leakage and becomes unusable. On the other hand, middle fiber breaks are known, in which only the plastic located between the fibers is damaged, as a result of which only the stiffness of the compressed gas tank decreases, but usually no leakage occurs. Fiber breaks usually only occur in the case of tank explosion, while middle fiber breaks can occur earlier, for example due to fatigue, without affecting the structural integrity of the tank. At the same time, middle fiber breaks can lead to cracks in the plastic matrix and correspondingly to leakage of the tank. In both cases, the service life and load capacity of the tank are reduced. Summary of the invention

[0004] Within the scope of the proposed invention, a diagnostic system, a tank system and a diagnostic method are proposed for determining the state of a compressed gas tank made of fiber-reinforced plastic. Further features and details of the invention are derived from the corresponding dependent claims, the description and the drawings. It is self-evident that the features and details described in connection with the diagnostic method according to the invention also apply in connection with the diagnostic system according to the invention or the tank system according to the invention, and vice versa, respectively, so that reference can always be made to each other with regard to the disclosure of the individual inventive aspects.

[0005] The proposed invention is used in particular for identifying a defective compressed gas tank as defective.

[0006] According to a first aspect of the proposed invention, a diagnostic system is proposed for ascertaining the state of a compressed gas tank made of fiber-reinforced plastic. The diagnostic system comprises an actuator, a sensor and an evaluation unit.

[0007] The excitation element is configured to introduce sound waves into the compressed gas tank.

[0008] The sensor is configured to detect sound waves introduced into the compressed gas tank by the excitation element.

[0009] The evaluation unit is configured to assign a characteristic value to the respective measured value ascertained by the sensor, which characteristic value describes the state of the compressed gas tank, and to output the assigned characteristic value to the output unit.

[0010] In the context of the proposed invention, an excitation element is understood to be a component that is able to introduce sound waves, in particular sound waves as structure-borne sound, into the compressed gas tank. The excitation element can be, for example, an ultrasonic transducer.

[0011] In the context of the proposed invention, a sensor is understood to be a component that is able to detect sound waves, in particular structure-borne sound, from a compressed gas tank. The sensor may be, for example, a knock sensor and / or a MEMS.

[0012] In the context of the proposed invention, an evaluation unit is understood to be a computing unit, in particular a control unit or a processor or any other programmable switching circuit.

[0013] The proposed invention is based on the principle that changes in sound waves introduced into a compressed gas tank (that is, into the housing body or the supporting structure of the compressed gas tank) are detected and the state of the compressed gas tank is inferred from the changes. The determined state is described by characteristic values ​​and output to an output unit such as a display or a memory. For example, changes in the tank are inferred by correlation of the recorded signals and general mathematical methods such as machine learning, for example, for anomaly detection or statistics.

[0014] Since sound waves extend or behave differently in fiber-plastic composites with existing core fiber breaks or fiber breaks than in composites with unbroken fibers or undamaged core fiber material, the state of the composite can be inferred from the behavior of the sound waves, in particular their changes.

[0015] The proposed diagnostic system can be reversibly arranged on a compressed gas tank and can be firmly connected to the compressed gas tank. In addition, the proposed diagnostic system can be configured to perform the proposed diagnostic method on the corresponding compressed gas tank once or multiple times. Accordingly, the diagnostic system can be used, for example, for diagnosing compressed gas tanks in a production line, or for diagnosing compressed gas tanks over time in, for example, a vehicle. In particular, the proposed diagnostic system can be configured to output an alarm signal when a defective compressed gas tank is identified, or to assign a characteristic value to the corresponding measured value, which corresponds to a defective state. Accordingly, the characteristic value corresponding to the defective state can include an alarm signal or can be an alarm signal.

[0016] It can be provided that the excitation element is configured to introduce structure-borne sound into the compressed gas tank and that the sensor is configured to detect the structure-borne sound emitted by the compressed gas tank.

[0017] In order to introduce structure-borne sound into the compressed gas tank or the housing of the compressed gas tank, the excitation element can be a contact element such as a vibration element, which repeatedly comes into contact with the compressed gas tank, for example strikes it, and thereby conditionally generates sound waves in the compressed gas tank or guides the sound waves by the compressed gas tank.

[0018] For detecting structure-borne sound passing through the compressed gas tank, a microphone or a knock sensor may be used, for example.

[0019] Furthermore, it can be provided that the analysis unit is configured to execute a machine learning algorithm which is trained to assign a first characteristic value or a second characteristic value to corresponding measured values ​​ascertained by the sensor, wherein the first characteristic value corresponds to a non-defective state and the second characteristic value corresponds to a defective state.

[0020] By means of a machine learning algorithm, such as an artificial neural network, for example, a corresponding characteristic value, such as, for example, "defective" or "non-defective", can be automatically assigned to a signal detected or measured by a sensor provided according to the invention. The machine learning algorithm can classify the corresponding signal accordingly. For this purpose, the evaluation unit can directly provide the signal as an input signal to the machine learning algorithm for use, or preprocess it by a certain number of preprocessing steps (such as, for example, a low-pass filter) and then provide it as an input signal to the machine learning algorithm for use.

[0021] It can be provided that the machine learning algorithm is trained on the basis of non-defective and / or defective compressed gas tanks and provided reference facts.

[0022] By training the machine learning algorithm on the basis of non-defective and / or defective compressed gas tanks and / or provided reference facts, a corresponding recognition limit of the machine learning algorithm can be set.

[0023] For example, it can be provided that the machine learning algorithm is trained in a so-called “unsupervised method” solely with the aid of non-defective compressed gas tanks, so that in the event of a signal deviation the machine learning algorithm automatically assigns a characteristic value to the corresponding compressed gas tank which describes the defective state without a model regarding the defective state.

[0024] Alternatively, it can be provided that the machine learning algorithm is trained in a so-called "supervised method" with reference facts, a certain number of non-defective compressed gas cylinders and a certain number of defective compressed gas cylinders, so that the machine learning algorithm develops a model of non-defective compressed gas cylinders and defective compressed gas cylinders. In this case, tolerances in terms of the quality of the compressed gas cylinders can be defined by the corresponding reference facts, so that the machine learning algorithm does not yet perform any assignment of characteristic values ​​that indicate a defective state even in the case of small deviations of the corresponding measurement signal from the expected signal.

[0025] Furthermore, it may be provided that the machine learning algorithm is pre-trained based on different structured material samples and reference facts in order to assign the corresponding material sample to a first category that maps a defective state or a second category that maps a non-defective state, and that the machine learning algorithm is validated with the aid of measured values ​​of at least one compressed gas tank.

[0026] By so-called "pre-training" or so-called "transfer learning" based on material samples, a large number of compressed gas cylinders for training the machine learning algorithm can be omitted, since the structural properties of defective or non-defective compressed gas cylinders can be described in a controlled manner from the material samples. By means of verification based on measurement data of at least one complete compressed gas cylinder, the quality of the transfer of the training can be determined or evaluated based on the material samples on the complete compressed gas cylinder.

[0027] In addition, it can be provided that the analysis unit is configured to implement a mathematical simulation model which simulates the intermediate fiber breaking strength and / or the load of a component of a lining of a corresponding compressed gas tank based on corresponding measured values ​​obtained by sensors, and obtains leakage caused by the simulated intermediate fiber breaking strength and / or the simulated load of the component of the lining, and the analysis unit is further configured to assign a corresponding characteristic value to the compressed gas tank based on the leakage obtained by the mathematical simulation model as an input signal for a machine learning algorithm.

[0028] A mathematical simulation model that assigns deviations in the measurement signal to a leak or simulates to what extent the deviations lead to a leak or do not lead to a leak can provide input signals or so-called "features" that can be used, for example, for training by a machine learning algorithm provided according to the invention. In this case, the simulation model can, in particular, enlarge the database (which is provided to the simulation model by experiments on compressed gas cylinders) by, for example, interpolation, so that the number of compressed gas cylinders required for determining the database for training the machine learning algorithm is minimized.

[0029] According to a second aspect, the proposed invention relates to a tank system comprising a compressed gas tank made of fiber-reinforced plastic and a possible configuration of the proposed diagnostic system.

[0030] The diagnostic system can be arranged, for example, on the compressed gas tank or in the area around the compressed gas tank, so that the excitation element can excite the compressed gas tank and the sensor can detect corresponding sound waves.

[0031] It can be provided that the actuating element is arranged at a first end of the compressed gas tank and that the sensor is arranged at a second end of the compressed gas tank opposite the first end.

[0032] The opposing arrangement of the excitation element and the sensor at the respective ends of the compressed gas tank has the effect that the sound waves introduced into the compressed gas tank by the excitation element propagate through the entire compressed gas tank or over the entire length of the compressed gas tank, thereby detecting changes or defects in the entire envelope structure of the compressed gas tank. In this case, a damping rubber suspension can be provided in order to suppress parasitic paths.

[0033] According to a third aspect, the proposed invention relates to a diagnostic method for ascertaining the state of a compressed gas tank made of fiber-reinforced plastic.

[0034] The diagnostic method comprises: introducing sound waves into the compressed gas tank by means of an excitation element; detecting the sound waves introduced into the compressed gas tank by the excitation element by means of a sensor; assigning a characteristic value to the corresponding measured value determined by the sensor, wherein the characteristic value indicates the state of the compressed gas tank; and outputting the characteristic value to an output unit.

[0035] The proposed diagnostic system is used, in particular, for carrying out the proposed diagnostic method.

[0036] It may be provided that the diagnostic method further comprises training a machine learning algorithm for assigning characteristic values ​​to the respective measured values ​​ascertained by the sensor, wherein the training comprises training the machine learning algorithm based on non-defective and / or defective compressed gas tanks and provided reference facts.

[0037] By using reference facts (i.e., correspondences with corresponding feature values ​​provided to the user, for example), the machine learning algorithm can continuously adjust its internal mathematical model in an iterative method until it ultimately results in a correspondence between training data and corresponding feature values ​​that is as consistent as possible with the reference facts.

[0038] In addition, it can be set that the diagnostic method includes: implementing a mathematical model, which determines the intermediate fiber breaking tension strength and / or load of the components of the lining (Innenfutter) of the corresponding compressed gas tank based on the measured values ​​obtained by the sensor, and the leakage caused by the simulated intermediate fiber breaking tension strength and / or the simulated load of the components of the lining; with the help of a machine learning algorithm, the characteristic value is assigned to the compressed gas tank, wherein the determined leakage is used as an input signal by the machine learning algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Further advantages, features and details of the invention are apparent from the following description, in which exemplary embodiments of the invention are described in detail with reference to the drawings. The features mentioned in the claims and in the description may be essential to the invention either individually or in any combination. The drawings show:

[0040] Figure 1 A schematic illustration shows a possible embodiment of the proposed tank system and a possible embodiment of the proposed diagnostic system,

[0041] Figure 2 A schematic illustration shows a possible embodiment of the proposed diagnostic method.

[0042] Figure 3 A flow chart of the training of a machine learning algorithm for carrying out one possible embodiment of the proposed diagnostic method is shown. DETAILED DESCRIPTION

[0043] exist Figure 1 2 , a tank system 100 is shown. The tank system 100 includes a compressed gas tank 101 and a diagnostic system 103 .

[0044] The diagnostic system 103 comprises an excitation element 105 in the form of an ultrasonic transducer, which is arranged at a mounting point 107 of the compressed gas tank 101 , a sensor 109 which is arranged at a mounting point 111 of the compressed gas tank 101 , and an analysis unit 113 which is communicatively connected to the excitation element 105 and the sensor 109 .

[0045] The sound waves 115 introduced into the compressed gas tank 101 by the excitation element 105 propagate on a transmission path along the compressed gas tank to the sensor 109 and are detected by the sensor 109. The measurement signal of the sensor 109 is transmitted to an evaluation unit 113.

[0046] The evaluation unit assigns characteristic values ​​such as “non-defective” or “defective” to the measurement signal using a machine learning algorithm, for example in the form of an artificial neural network, and outputs the characteristic values ​​to an output unit (not shown), for example as a warning signal.

[0047] exist Figure 2 It is shown in detail in Figure 1 The excitation pulse 201 generates a sound wave 203 which is introduced into a composite structure 205 of the compressed gas tank consisting of fibers 207 (such as carbon fibers, for example) and plastic 209 (such as epoxy resin, for example). If the composite structure 205 is free of defects, the sound wave 203 can pass through the plastic 209 essentially unhindered and cause a characteristic measurement signal 211.

[0048] In the event of a defect in composite structure 213 , the propagation of sound wave 203 is blocked or delayed by composite structure 213 , for example by crack 215 , so that a measurement signal 217 that differs from characteristic measurement signal 211 is generated.

[0049] exist Figure 3 , a method 300 for training a first machine learning algorithm is shown. The method 300 begins with a classification process 301, in which the first machine learning algorithm is trained on the basis of material samples with different structural properties. In this case, the measured values ​​obtained from the material samples can be used in order to generate input data or so-called "features" in a preprocessing step 303, which are provided to the first machine learning algorithm for the classification process, in which the first machine learning algorithm assigns the characteristic value "defective" or the characteristic value "non-defective" to the corresponding material sample. The correspondingly trained first machine learning algorithm is cached in a storage step 305 in order to be available for a migration step 311. By using material samples to train the first machine learning algorithm, the use of compressed gas cylinders can be minimized.

[0050] Alternatively, a regression can be carried out based on the measured values, which mathematically maps a large number of intermediate states accordingly.

[0051] Furthermore, method 300 includes a further training step 307 , in which a further machine learning algorithm is trained using measured values ​​ascertained with the aid of a complete compressed gas tank.

[0052] In order to minimize the number of compressed gas tanks to be used for the further training step 307, in an optional modeling step 309, the measured values ​​ascertained in the further training step 307 can be used to generate a mathematical simulation model which simulates the intermediate fiber breaking tensile strength or intermediate fiber breaking load of a material sample, in particular a tank, and assigns a value to it for a corresponding leak. Accordingly, the database describing the behavior of a complete compressed gas tank can be enlarged by the simulation model, for example by interpolating between the corresponding measured values ​​or by ascertaining additional values.

[0053] In the migration step 311, the mathematical model that serves as the basis for the first machine learning algorithm is merged with the mathematical model that serves as the basis for another machine learning algorithm, or the first machine learning algorithm is trained according to the values ​​obtained by the simulation model, so that the first machine learning algorithm is verified by the measured values ​​of the complete compressed gas tank.

[0054] In an application step 313 , the resulting mathematical model is used in a diagnostic system with the aid of a first machine learning algorithm in order to diagnose a state or so-called “health state” of the compressed gas tank.

Claims

1. A diagnostic system (103) for determining the state of a compressed gas tank (101) made of fiber-reinforced plastic, wherein: The diagnostic system (103) comprises: - an excitation element (105), - a sensor (109), - an analysis and processing unit (113), wherein the excitation element (105) is configured to introduce sound waves into the compressed gas tank (101), The sensor (109) is configured to detect sound waves introduced into the compressed gas tank (101) by the excitation element (105), The evaluation unit (113) is configured to assign a characteristic value to a corresponding measured value ascertained by the sensor (109), the characteristic value indicating the state of the compressed gas tank (101), and to output the assigned characteristic value to an output unit.

2. The diagnostic system (103) according to claim 1, characterized in that The excitation element (105) is configured to introduce structure-borne sound into the compressed gas tank (101), and the sensor (109) is configured to detect the structure-borne sound emitted by the compressed gas tank (101).

3. The diagnostic system (103) according to claim 1 or 2, characterized in that: The analysis unit (113) is configured to implement a machine learning algorithm, which is trained to assign a first characteristic value or a second characteristic value to a corresponding measured value obtained by the sensor (109), wherein the first characteristic value corresponds to a non-defective state and the second characteristic value corresponds to a defective state.

4. The diagnostic system (103) according to claim 3, characterized in that: The machine learning algorithm is trained based on non-defective and / or defective compressed gas cylinders (101) and / or provided benchmark facts.

5. The diagnostic system (103) according to claim 4, characterized in that The machine learning algorithm is pretrained with the aid of differently structured material samples and reference facts in order to assign the respective material sample to a first class representing a defective state or to a second class representing a non-defective state, and is validated with the aid of measured values ​​of at least one compressed gas tank.

6. The diagnostic system (103) according to claim 4 or 5, characterized in that: The analysis unit (113) is configured to implement a mathematical simulation model, which simulates the intermediate fiber breaking tension and / or the load of the lining of the corresponding compressed gas tank (101) based on the corresponding measured values ​​determined by the sensor (109), and determines the leakage caused by the simulated intermediate fiber breaking tension and / or the simulated load of the lining, and, The evaluation unit (113) is further configured to assign corresponding characteristic values ​​to the compressed gas tank (101) as a function of the values ​​ascertained by the mathematical simulation model as input signals for the machine learning algorithm.

7. A tank system (100), wherein: The tank system comprises: - a compressed gas tank (101) made of fiber-reinforced plastic, - A diagnostic system (103) according to any one of claims 1 to 6.

8. The tank system (100) according to claim 7, characterized in that The actuating element (105) is arranged on a first end of the compressed gas tank (101), and the sensor (109) is arranged on a second end of the compressed gas tank (101) opposite the first end.

9. A diagnostic method for determining the condition of a compressed gas tank (101) made of fiber-reinforced plastic, wherein: The diagnostic method comprises: - introducing sound waves into the compressed gas tank (101) by means of an excitation element (105), - detecting, by means of a sensor (109), the sound waves introduced into the compressed gas tank (101) by the excitation element (105), - assigning a characteristic value to the respective measured value ascertained by the sensor (109), wherein the characteristic value describes the state of the compressed gas tank (101), - outputting the characteristic value to an output unit.

10. The diagnostic method according to claim 9, characterized in that The diagnostic method further comprises: - training (300) a machine learning algorithm in order to assign characteristic values ​​to the corresponding measured values ​​ascertained by the sensor (109), Therein, the training (300) includes training the machine learning algorithm based on non-defective and / or defective compressed gas tanks and provided benchmark facts.

11. The diagnostic method according to claim 10, characterized in that: The diagnostic method comprises: - implementing (303) a mathematical simulation model which determines the intermediate fiber breaking tension and / or the load of the lining of the corresponding compressed gas tank (101) based on the measured values ​​determined by the sensor (109) and the leakage caused by the simulated intermediate fiber breaking tension and / or the simulated load of the lining, - assigning (313) the characteristic value to the compressed gas tank (101) by means of the machine learning algorithm, wherein the ascertained leakage is used as an input signal by the machine learning algorithm.

Images