Noise analyzer
The abnormal noise analysis device uses a machine-learned model to compare frequency spectra, improving detection accuracy by accounting for combustion pressure fluctuations.
Patent Information
- Application Number
- JP2023080701
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-16
- Publication Date
- 2024-11-28
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing methods for detecting abnormal noise in internal combustion engines are affected by variations in combustion pressure, leading to potential undetection of non-periodic abnormal noises.
An abnormal noise analysis device utilizing a learned model subjected to machine learning, which outputs a frequency spectrum based on combustion pressure, compares actual and reference frequency spectra to determine abnormal noise occurrence.
Enhances the accuracy of detecting abnormal noise in internal combustion engines by mitigating the impact of combustion pressure variations.
Smart Images

Figure 2024164951000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to an abnormal noise analysis device that determines whether or not an abnormal noise occurs in an internal combustion engine. [Background technology]
[0002] Patent Document 1 discloses an abnormal sound analysis device that judges knocking in an internal combustion engine. The abnormal sound analysis device identifies periodic abnormal sounds that occur periodically during engine operation by comparing the abnormal sound that occurs in the most recent cycle with abnormal sounds that occur in cycles earlier than the most recent cycle during engine operation. The abnormal sound analysis device then calculates the volume of abnormal sounds other than periodic abnormal sounds after removing the periodic abnormal sounds from the sounds that occur during engine operation. [Prior art documents] [Patent documents]
[0003] [Patent Document 1] Patent No. 6491050 Summary of the Invention [Problem to be solved by the invention]
[0004] When the engine is operating, there is a large variation in combustion pressure, so the above-mentioned method of comparing the most recent abnormal noise with abnormal noises that occurred in an earlier period is affected by the variation in combustion pressure, and there is a risk that abnormal noises other than periodic abnormal noises will not be detected. [Means for solving the problem]
[0005] The abnormal sound analysis device for solving the above problem includes a storage device that stores a trained model that has been subjected to machine learning, and an execution device. The trained model outputs, as an output variable, a frequency spectrum that indicates the relationship between the sound pressure and frequency of a sound generated in the internal combustion engine when a combustion pressure of the internal combustion engine is input as an input variable. The execution device executes the following operations during engine operation: measuring the combustion pressure of the internal combustion engine and the sound generated in the internal combustion engine, acquiring, as a real frequency spectrum, a frequency spectrum that indicates the relationship between the sound pressure and frequency of the measured sound generated in the internal combustion engine, acquiring, as a reference frequency spectrum, the frequency spectrum that is the output variable of the trained model when the measured combustion pressure of the internal combustion engine is used as the input variable of the trained model, and determining whether or not an abnormal sound has occurred in the internal combustion engine by comparing the real frequency spectrum with the reference frequency spectrum. Effect of the Invention
[0006] The abnormal sound analysis device described above has the effect of improving the accuracy of determining whether or not an abnormal sound has occurred in an internal combustion engine. [Brief description of the drawings]
[0007] [Figure 1] FIG. 1 is a diagram showing the configuration of an abnormal sound analysis device according to an embodiment and a schematic configuration of an internal combustion engine to be analyzed. [Diagram 2] FIG. 2 is a flowchart showing a series of processes executed by the abnormal sound analysis device of FIG. [Diagram 3] FIG. 3 is a graph comparing the actual frequency spectrum with the reference frequency spectrum. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0008] Hereinafter, one embodiment of an abnormal sound analysis device will be described with reference to Figs. <Internal combustion engine> As shown in Fig. 1, an internal combustion engine 10 includes a cylinder 11 and a piston 12 that moves up and down within the cylinder 11. The piston 12 is connected to a crankshaft 14 via a connecting rod 13. Therefore, when the piston 12 moves up and down within the cylinder 11, the crankshaft 14 rotates in conjunction with the up and down movement of the piston 12.
[0009] Air is introduced into a combustion chamber 15 defined by the piston 12 in the cylinder 11 via an intake passage 16. In the combustion chamber 15, an air-fuel mixture containing air and fuel injected from a fuel injection valve 17 is burned. Exhaust gas generated by the combustion of the air-fuel mixture is discharged from the combustion chamber 15 to an exhaust passage 18.
[0010] <Abnormal sound analysis device> The abnormal sound analysis device 30 receives signals from a microphone 21 that collects sounds generated in the internal combustion engine 10, a combustion pressure sensor 22 that detects the combustion pressure of the internal combustion engine 10, and the like.
[0011] An example of the abnormal sound analysis device 30 is an electronic control device. In this case, the abnormal sound analysis device 30 includes a CPU 31 and a first storage device 32. The CPU 31 corresponds to an "execution device." The first storage device 32 stores a control program executed by the CPU 31. The CPU 31 executes the control program to determine whether or not an abnormal sound has occurred in the internal combustion engine 10.
[0012] The abnormal sound analysis device 30 includes a second storage device 33. The second storage device 33 corresponds to a "storage device." The second storage device 33 stores a trained model LM that has been subjected to machine learning. The trained model LM is a model that outputs, as an output variable, a frequency spectrum that indicates the relationship between the sound pressure and frequency of the sound generated in the internal combustion engine 10 when the combustion pressure of the internal combustion engine 10 is input as an input variable. The trained model LM has been trained so that, when the combustion pressure is input as an input variable, the trained model LM can output, as an output variable, a frequency spectrum in the case where the internal combustion engine 10 does not generate an abnormal sound when the combustion pressure is generated in the internal combustion engine 10. The frequency spectrum output from the trained model LM is referred to as a "reference frequency spectrum SPB." The trained model LM is constructed, for example, by a neural network.
[0013] <Abnormal sound analysis processing> 2 and 3, an abnormal sound analysis process which is a series of processes executed by the CPU 31 to determine whether or not an abnormal sound has occurred in the internal combustion engine 10 will be described. Fig. 2 is a flowchart showing the abnormal sound analysis process.
[0014] In step S11, the CPU 31 simultaneously measures the combustion pressure of the internal combustion engine 10 and the sound generated by the internal combustion engine 10. In the following step S13, the CPU 31 performs known FFT processing on the sound measured in step S11. In the next step S15, the CPU 31 acquires the data subjected to the FFT processing as a real frequency spectrum SPR. The real frequency spectrum SPR is a frequency spectrum indicating the relationship between the sound pressure and frequency of the sound measured in the processing of step S11.
[0015] In step S17, the CPU 31 acquires a reference frequency spectrum SPB which is an output variable of the learned model LM when the combustion pressure measured in the process of step S11 is input as an input variable of the learned model LM.
[0016] Then, in step S19, the CPU 31 compares the real frequency spectrum SPR with the reference frequency spectrum SPB to determine whether or not an abnormal noise has occurred in the internal combustion engine .
[0017] Fig. 3 shows an example of a comparison between the real frequency spectrum SPR and the reference frequency spectrum SPB. In Fig. 3, the solid line indicates the real frequency spectrum SPR, while the dashed line indicates the reference frequency spectrum SPB. In the example shown in Fig. 3, in a certain frequency range, a deviation occurs between the sound pressure in the real frequency spectrum SPR and the sound pressure in the reference frequency spectrum SPB.
[0018] The CPU 31 calculates the deviation ΔSP between the sound pressure in the real frequency spectrum SPR and the sound pressure in the reference frequency spectrum SPB for each frequency. If there is a frequency at which the deviation ΔSP is equal to or greater than a predetermined threshold, the CPU 31 determines that an abnormal sound has occurred in the internal combustion engine 10. On the other hand, if the deviation ΔSP is not equal to or greater than the threshold at any frequency, the CPU 31 determines that an abnormal sound has not occurred in the internal combustion engine 10. After making this abnormal sound determination, the CPU 31 ends the abnormal sound analysis process shown in FIG. 2.
[0019] <Actions and Effects of the Present Embodiment> The abnormal sound analysis device 30 judges whether an abnormal sound has occurred in the internal combustion engine 10 by comparing a reference frequency spectrum SPB corresponding to the sound pressure actually generated in the internal combustion engine 10 with the actual frequency spectrum SPR. This allows the comparison to be performed without taking into account variations in the combustion pressure. Therefore, the abnormal sound analysis device 30 can increase the accuracy of judging whether an abnormal sound has occurred in the internal combustion engine 10. [Explanation of symbols]
[0020] 10...internal combustion engine, 30...abnormal sound analysis device, 31...CPU, 33...second storage device, LM...trained model.
Claims
[Claim 1] The present invention provides a method for learning a machine learning model by using a memory device that stores a trained model having machine learning functions, and a computer-implemented execution device. The trained model outputs, when a combustion pressure of an internal combustion engine is input as an input variable, a frequency spectrum indicating a relationship between a sound pressure and a frequency of a sound generated in the internal combustion engine as an output variable; The execution device is Measuring a combustion pressure of the internal combustion engine and a sound generated by the internal combustion engine during engine operation; Obtaining a frequency spectrum indicating a relationship between a sound pressure and a frequency of the sound generated in the internal combustion engine as a real frequency spectrum; acquiring, as a reference frequency spectrum, the frequency spectrum which is the output variable of the trained model when the measured combustion pressure of the internal combustion engine is used as the input variable of the trained model; and determining whether or not an abnormal noise occurs in the internal combustion engine by comparing the actual frequency spectrum with the reference frequency spectrum. Abnormal sound analysis device.
Citation Information
Patent Citations
Sensor for counting number of active bacteria
JP1989091050A
