Computer-implemented method for detecting anomalies in the use of a machine

Abstract

The invention relates to a computer-implemented method for detecting anomalies in the use of a machine, - Acquisition of sensor data (S10) with a smart sensor as a time series; - Determining time domain features (S12) from the sensor data; - Transforming the acquired time series (S14) into a frequency domain; - Determining frequency domain features (S16) from the sensor data; - evaluating the provided features (S18) by an evaluation algorithm; and - Determining the presence of an anomaly (S20) from the result of the evaluation of the features, wherein the smart sensor comprises an acceleration sensor and wherein the sensor data comprises measured values ​​of the acceleration of the machine in at least one spatial direction, wherein the time domain features comprise the root mean square, the mean value, the crest factor and the peak-to-peak value of the acquired time series, wherein the frequency domain features comprise the root mean square of a freely defined window frequency, the root mean square of a defined window frequency, the spectral flatness, the N-Max peak and the number of zero crossings within the frequency spectrum.

Description

[0001] The invention relates to the analysis of sensor data, in particular data from a vibration or structure-borne sound sensor in machines, in particular with rotating elements. State of the art

[0002] A diagnostic module is known from WO 2013 072 145 A2. This diagnostic module is intended for machines, particularly hydraulic displacement machines. It combines an integrated vibration sensor with additional sensors such as temperature and pressure sensors. This integration enables compact monitoring of the entire device status. The diagnostic module can be operated wirelessly and is particularly suitable for rotating devices such as hydraulic displacement machines. It also enables intelligent diagnostics and prognosis by processing the measurement data using internal software. Placing the diagnostic module on or in the rotating machine facilitates its use.

[0003] However, the diagnostic module lacks a CAN interface and only features a vibration sensor. These are therefore simple, intelligent sensors for condition monitoring on rotating machines with conventional signal processing, such as using fast Fourier transforms (FFTs) and alarm thresholds. Other features are generally not considered. The familiar diagnostic module therefore has limited applications.

[0004] The effort required for commissioning and parameterization, as well as for any necessary parameter adjustments, for example, to set new alarm thresholds, can be very high and often requires expert knowledge. For example, monitoring rolling bearings requires knowledge of the exact geometry—that is, the diameter of the raceways and rolling elements, as well as the number of rolling elements. For variable-speed applications, a speed signal must be available to track the excitation orders.

[0005] To monitor a wide variety of fault types, numerous frequency bands must be defined for each component. These include, for example, the fundamental frequency of the respective component, higher harmonics, sidebands, and others.

[0006] The condition monitoring function integrated in intelligent vibration sensors by evaluating the vibration time signal using a characteristic value formation with subsequent transformation into the frequency domain, for example via a fast Fourier transformation, a feature formation and a trend display with alarm threshold monitoring is therefore known and already integrated in various intelligent vibration sensors.

[0007] Against this background, the object of the invention is to create an efficient method for anomaly monitoring with a smart sensor that can be maintained or updated with reduced effort.

[0008] The problem is solved by the subject matter of the independent claims. Disclosure of the invention

[0009] According to a first aspect of the invention, this object is achieved by a computer-implemented method for detecting anomalies in the use of a machine, - Collecting sensor data with a smart sensor as a time series; - Determining time-domain features from the sensor data; - Transforming the recorded time series into a frequency domain; - Determining frequency domain features from the sensor data; - Evaluating the provided features using an evaluation algorithm; and - Determine the presence of an anomaly from the result of the evaluation of the features.

[0010] The smart sensor includes an acceleration sensor and the sensor data includes measured values ​​of the acceleration of the machine in at least one spatial direction.

[0011] Preferably, the machine's acceleration is measured in two, particularly preferably three, spatial directions. The more data collected, the greater the resource requirements for processing it. However, with the amount of data, the possibility of detecting an anomaly also increases. This increase is not necessarily linear, so a choice must be made regarding how much and what type of data should be collected and how much resource consumption is acceptable for processing.

[0012] The time-domain features include the root mean square, the mean value, the crest factor, and the peak-to-peak value of the acquired time series. The frequency-domain features include a user-defined window frequency, a defined window frequency, the spectral flatness, the N-max peak, and the number of zero crossings within the frequency spectrum.

[0013] Vibration-based condition monitoring of machines with rotating parts is based on the analysis of vibrations, also known as structure-borne noise, that occur during machine operation. This method uses vibration sensors to measure and evaluate the mechanical vibrations of machines, particularly their rotating parts. The idea behind it is that changes in vibration behavior can indicate potential problems or anomalies in the machine's operating condition.

[0014] A vibration sensor can be mounted at strategic locations on or inside the machine. For example, it can be on the housing, especially near rotating or moving components. The sensors convert the structure-borne noise of the machine into electrical signals. The positioning of the sensors is therefore crucial to ensure representative recording of the machine vibrations.

[0015] The signals generated by the sensors are recorded as time series in real time or at regular intervals. The raw data can be transformed into the frequency domain using signal processing techniques, such as Fourier transformation. This enables a detailed analysis of the various frequency components of the vibrations.

[0016] Frequency analysis makes it possible to identify characteristic patterns in vibration behavior. Each machine generates characteristic frequencies due to its specific components and operating conditions. These frequencies can be associated with the machine's normal operating conditions.

[0017] Deviations from expected vibration patterns are considered anomalies. These anomalies may indicate wear, imbalance, alignment problems, bearing problems, or other malfunctions in the machine.

[0018] Vibration-based condition monitoring enables early detection of developing problems, which can lead to improved machine availability, longer component life, and reduced operating costs.

[0019] To implement the proposed method, both time domain and frequency domain features are used for evaluation.

[0020] The "root mean square" refers to a statistical metric that measures the average square deviation of a set of values ​​from their mean. It is also known as the standard deviation or root mean square (RMS).

[0021] The formula for the root mean square (RMS) is: RMS=1n∑i=1Nxi2, where x iwhere the individual values ​​are the time series and N is the number of values. The mean of the squared deviations is calculated, and then the square root of the result is taken.

[0022] The root mean square is widely used in various applications, especially in signal processing. In signal processing, it is often used to measure the effect of fluctuations or variations in a signal, with larger deviations being given greater weight.

[0023] The crest factor is a term in electrical engineering that describes the ratio of the peak value to the effective value of an alternating quantity. It is used primarily in various fields such as electrical measurement technology, communications engineering, sound engineering, and acoustics. The crest factor (k s ) is defined as the ratio of the maximum amount (|X| max ) of the alternating quantity X to the effective value (X eff ): ks=|X|maxXeff.

[0024] The crest factor is used to roughly describe the waveform of an alternating quantity and serves as a characteristic value. The square of the crest factor is called the peak-to-average power ratio (PAPR) and indicates the ratio of peak power to average power of a signal. This is usually measured in decibels (dB).

[0025] The "peak-to-peak value" refers to the maximum value of an alternating quantity over a specific period of time, measured from the highest peak to the lowest peak. This value is also called the "peak-to-peak value" and indicates the total deflection or range of variation of the alternating quantity.

[0026] In the context of acoustic vibrations, and especially structure-borne sound, the peak-to-peak value represents the distance between the highest positive and lowest negative points of the vibration within a given time frame. Mathematically speaking, the peak-to-peak value corresponds to the difference between the maximum positive value and the minimum negative value.

[0027] The peak-to-peak value is important for understanding the full range of variation of a quantity. This makes the peak-to-peak value a valuable feature for anomaly analysis.

[0028] The frequency spectrum can be determined in particular by a fast Fourier transform (FFT).

[0029] Higher frequencies, especially those in the third or octave bands, are particularly informative.

[0030] Another frequency-domain characteristic used is spectral flatness. Spectral flatness refers to the even distribution of frequencies in a signal or spectrum. In a flat spectrum, all frequencies are represented with similar intensity, without any particular frequencies dominating. A signal with high spectral flatness therefore has an even energy distribution across the frequency range. This property is important in audio engineering, for example, to ensure that a sound signal sounds natural and balanced without certain frequencies being overemphasized.

[0031] N-Max Peak refers to a specific number (N) of maximum amplitudes whose frequencies are recorded. For example, the frequencies of the four largest amplitudes can be recorded if N = 4.

[0032] The advantage of the described method and the resulting combination of time-domain and frequency-domain features is that no adjustments, parameterization, or adaptation are necessary during commissioning. This saves the commissioning engineer or customer considerable time and significantly reduces costs.

[0033] In one embodiment, the smart sensor comprises a gyroscope, wherein the sensor data includes the position of the machine in space.

[0034] Recording position data using a gyroscope is a simple means of obtaining additional useful information about the condition of a machine. The data provided by the gyroscope is particularly informative when the machine or the working part is moving and the data is analyzed in conjunction with the data from the accelerometer. Depending on the machine's position, the accelerometer data may need to be evaluated differently along the measurement axes. Combining the accelerometer data with the data from a gyroscope can at least partially compensate for or even offset this position dependency.

[0035] In one embodiment, the sensor data further comprise measured values ​​of a tachometer in the machine, the pressure, in particular the working pressure in a hydraulic sub-system of the machine, the temperature of the machine and / or a part thereof, a manipulated variable and / or a state variable.

[0036] Integral process variables of the machine, such as the working pressure in a hydraulic cylinder and others, can also be used to detect an anomaly early on. It is rare for an anomaly to manifest itself in only one process variable. Therefore, adding these process variables can increase the probability of detecting an anomaly early on.

[0037] In one embodiment, in particular, two temperature sensors can be used, the data of which are used for anomaly detection.

[0038] A temperature sensor can be mounted near the bottom of the machine's housing. This temperature sensor is not heated by the machine's electronics and can therefore measure the temperature of the machine or part of the machine being monitored independently of the electronics temperature.

[0039] A second temperature sensor can be provided to measure the temperature of the electronics and thus simultaneously serve as a protection for the electronics. If the temperature of the electronics exceeds a certain value, it can be shut down, particularly through an emergency shutdown process, thus protecting it from damage caused by excessive temperatures.

[0040] Both positions are also accessible in versions with only one temperature sensor.

[0041] In one embodiment, the features are transmitted to an edge cloud for evaluation.

[0042] The transmission of the features to an edge cloud can be carried out in particular via a CAN interface of the smart sensor.

[0043] The advantage of transmitting the features to an edge cloud is that the evaluation can be handled by the cloud system, thus reducing the system requirements for the smart sensor. Since the smart sensor does not have to perform the evaluation independently, the RAM and computing capacity, in particular, can be smaller.

[0044] In the edge cloud, complex analyses based on models with more variables can be performed due to the potentially more resource-intensive hardware.

[0045] In one embodiment, the evaluation is carried out by a computing unit of the machine itself or by a component of the smart sensor.

[0046] Carrying out the evaluation by the machine or a component of the smart sensor has the advantage that no data needs to be transmitted.

[0047] Avoiding data transmission prevents sensitive data in particular from being intercepted and spied on by unauthorized third parties. This ensures a high level of data security.

[0048] Furthermore, regardless of the ability to receive and transmit data—that is, regardless of the transmission power—the machine can process the data independently, detect anomalies, and initiate corrective or protective measures if necessary. If the machine, such as a construction machine, is outside the transmission and reception range of the system that normally performs the analysis, the machine can still initiate the necessary measures.

[0049] In one embodiment, the evaluation comprises processing the features by a machine learning algorithm.

[0050] Processing can be performed in the edge cloud for computationally intensive evaluations and procedures or on-board the device for fast, streamlined evaluation and indication. More complex procedures can be calculated locally, for example.

[0051] A machine learning algorithm is an algorithm designed to automatically detect patterns and relationships in data and make predictions or decisions. It is created by training on existing data and can then be applied to new, unknown data to generate predictions or classifications.

[0052] A machine learning algorithm can take various forms, such as linear models, decision trees, support vector machines, neural networks, and many others. It is optimized by learning from the training data, identifying patterns and rules to make the best possible predictions or classifications for new data.

[0053] The effectiveness of a machine learning algorithm depends on several factors, including the quality and quantity of training data, the choice of model, the model configuration, and the evaluation of the model using evaluation metrics. The model can be continuously improved and optimized to maximize accuracy and performance.

[0054] The linear regression model assumes a linear relationship between a dependent variable and one or more independent variables and is used to make predictions about continuous values.

[0055] Support vector machines (SVMs) are a model used for classification or regression that detects patterns in the data. They search for the optimal separation between different classes or attempt to fit a continuous function to the data.

[0056] Decision trees are a model that creates decision rules in the form of a tree diagram. They divide data based on features and enable predictions or classifications.

[0057] A probabilistic model is Naive Bayes, which is based on Bayes' theorem and is used for classification. It assumes that features are independent of each other and calculates the probability of a particular class based on the given features.

[0058] Neural networks refer to models that are primarily used for processing a wide variety of data. The architecture of a neural network comprises multiple nodes, neurons, or nodes, arranged in layers.

[0059] The detection of anomalies in data can be treated as a classification task, with the algorithm attempting to classify the smart sensor data. If the data is not assigned to a class corresponding to nominal operation, it can be inferred that an anomaly is present. The anomalies can, in turn, be divided into different classes, representing, for example, different error or fault conditions.

[0060] Anomaly detection can also be treated as a regression task. Regression tasks examine and identify trends in data. For example, if the vibrations of a machine increase without a change in operating mode, this may indicate a malfunction, such as increasing wear or an abnormal operating temperature due to increased friction. However, the exact assignment of a trend to a specific fault fundamentally depends on the machine and the monitored process.

[0061] In embodiments, the machine learning algorithm may be configured to first solve a regression task and then a classification task, wherein the result of the regression task is used as a parameter in solving the classification task.

[0062] In one embodiment, the machine performs a response when an anomaly is detected in the sensor data.

[0063] The reaction may, for example, include switching off the machine.

[0064] This advantageously enables an emergency stop system. The emergency stop system can prevent the machine from performing the process even when an anomaly is present and potentially incorrect processes are being performed. Specifically, it could be used to monitor machine wear, so that if excessive wear occurs, the machine can be readjusted, repaired, or replaced. For processes involving high-value workpieces, for example, made of expensive materials, costs are saved because fewer processes are performed incorrectly.

[0065] In another example, the response may be to generate an alert signal when an anomaly is detected.

[0066] Advantageously, generating a warning signal can alert an employee operating or monitoring the machine to the anomaly. This allows the cause of the anomaly to be identified and remedied early on, without potentially causing the machine's work process to continue to be executed incorrectly and the machine to perform its work unnecessarily or even harmfully.

[0067] Preferably, the machine can provide the data used to detect the anomaly or the type of anomaly to facilitate troubleshooting for maintenance. This may provide a specific indication of a fault source and / or an assessment of the urgency for maintenance or repair.

[0068] To further improve maintenance or repair, the machine or smart sensor can store the acquired sensor data and, if necessary, send it, possibly with a timestamp, to a central evaluation system, such as an edge cloud. In this embodiment, the machine or smart sensor can be configured to detect an anomaly in the sensor data, while the evaluation system performs a more detailed analysis of the data only upon a corresponding command. The evaluation system can preferably be equipped with more powerful hardware, allowing even more complex analysis methods to be performed in software.

[0069] In one embodiment, the sensor data is stored in a cache, wherein the sensor data stored in the cache is transferred to a protected memory when an anomaly is detected.

[0070] In this embodiment, the sensor data can be protected from data loss. Some anomalies may cause damage to the machine or its components, including the smart sensor. In these cases, the proposed method can be used to retrospectively identify the cause of the destructive anomaly. This may prevent these anomalies from occurring in future machine operations.

[0071] In one embodiment, the buffer may be a ring buffer.

[0072] The advantage of ring buffers is that they can cache data very efficiently and overwrite it automatically. This avoids complex storage management or even the manual replacement of data storage devices and simplifies the retention of data relevant to an anomaly.

[0073] The ring buffer should be sized to provide sufficient data for detecting anomalies and, if necessary, sufficient data for a more complex analysis of the anomaly causes. Sufficient storage space should be reserved, especially for detecting long-term trends. The absolute amount of storage required should be adapted to the machine and the work processes it performs.

[0074] In a further aspect, the invention relates to a computer program with program code for carrying out a method as described above when the computer program is executed on a computer.

[0075] In a further aspect, the invention relates to a computer-readable data carrier with program code of a computer program for carrying out a method as described above when the computer program is executed on a computer.

[0076] In a further aspect, the invention relates to a system for detecting anomalies during the use of a machine, wherein the system is designed to carry out a method as described above, wherein the system comprises at least one computing unit and at least one smart sensor communicatively connected to the computing unit.

[0077] In a further aspect, the invention relates to a machine with a system for detecting anomalies in the use of the machine as described above.

[0078] In summary, the present invention provides a method for detecting anomalies in the use of a machine, a computer program with program code, a computer-readable data carrier, a system for detecting anomalies in the use of a machine, and a corresponding machine.

[0079] The described designs and further training courses can be combined as desired.

[0080] Further possible embodiments, developments and implementations of the invention also include combinations of features of the invention described previously or below with regard to the exemplary embodiments that are not explicitly mentioned. Short description of the drawings

[0081] The accompanying drawings are intended to provide a further understanding of embodiments of the invention. They illustrate embodiments and, in conjunction with the description, serve to explain principles and concepts of the invention.

[0082] Other embodiments and many of the aforementioned advantages will become apparent upon review of the drawings. The elements shown in the drawings are not necessarily drawn to scale.

[0083] They show: Fig. 1 schematically shows the sequence of the method according to an embodiment of the invention.

[0084] In the figure of the drawing, the same reference symbols designate the same or functionally identical elements, parts or components, unless otherwise stated.

[0085] Fig. 1 schematically shows the sequence of a method according to an embodiment.

[0086] The method begins in step S10 with the acquisition of sensor data. The sensor data is acquired by a smart sensor, which includes at least one acceleration sensor. In embodiments, the smart sensor may include other sensor types, such as a gyroscope, a thermometer, or a microphone for detecting sound.

[0087] The sensor data is acquired as a time series, from which time-domain features are extracted in step S12. The time-domain features include the root mean square, the mean, the crest factor, and the peak-to-peak value of the time series. Additional time-domain features, such as the standard deviation or other error information, can also be determined.

[0088] In step S14, the time series is transformed into the frequency domain. This can be done, for example, using a fast Fourier transform (FFT). Other transformation methods, such as wavelet transform, are also possible.

[0089] In step S16, frequency features are determined from the transformed time series. The features include the root mean square of a window frequency freely defined, for example, by a user or a machine-specific frequency, the root mean square of a defined window frequency, in particular independent of the machine's application, the spectral flatness, the N-max peak, and the number of zero crossings.

[0090] In step S18, the features, i.e., the time-domain features and the frequency-domain features, are evaluated by an evaluation algorithm. The evaluation algorithm can be implemented, for example, as a machine learning algorithm. Furthermore, the evaluation algorithm can be executed by the smart sensor or by an edge cloud communicatively connected to the machine or the smart sensor, in which an evaluation and learning method (e.g., a neural network) can perform a more in-depth analysis.

[0091] In step S20, the evaluation algorithm determines whether an anomaly exists or whether the machine is in a designated operating state. An anomaly can be detected, for example, by an unusual vibration in moving parts of the machine. In embodiments, the type of anomaly can also be determined. QUOTES CONTAINED IN THE DESCRIPTION

[0000] This list of documents submitted by the applicant was generated automatically and is included solely for the convenience of the reader. This list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions. Cited patent literature

[0000] WO 2013 072 145 A2

[0002]

Claims

[1] Computer-implemented method for detecting anomalies in the use of a machine, - Acquisition of sensor data (S10) with a smart sensor as a time series; - Determining time domain features (S12) from the sensor data; - Transforming the acquired time series (S14) into a frequency domain; - Determining frequency domain features (S16) from the sensor data; - evaluating the provided features (S18) by an evaluation algorithm; and - Determining the presence of an anomaly (S20) from the result of the evaluation of the features, wherein the smart sensor comprises at least one acceleration sensor and wherein the sensor data comprise measured values ​​of the acceleration of the machine in at least one spatial direction, where the time-domain features include the root mean square, the mean, the crest factor and the peak-to-peak value of the acquired time series wherein the frequency domain features include the root mean square of a freely defined window frequency, the root mean square of a defined window frequency, the spectral flatness, the N-Max peak and the number of zero crossings within the frequency spectrum. [2] A computer-implemented method according to any one of the preceding claims, wherein the smart sensor comprises a gyroscope and wherein the sensor data comprises the position of the machine in space. [3] Computer-implemented method according to one of the preceding claims, wherein the sensor data further comprise measured values ​​of a tachometer in the machine, the pressure, in particular the working pressure in a hydraulic sub-system of the machine, the temperature of the machine and / or a part thereof, a manipulated variable and / or a state variable. [4] Computer-implemented method according to one of the preceding claims, wherein the features are transmitted to an edge cloud for evaluation. [5] Computer-implemented method according to one of claims 1 to 4, wherein the evaluation is carried out by a computing unit of the machine itself or by a component of the smart sensor. [6] Computer-implemented method according to one of the preceding claims, wherein the evaluation comprises processing of the features by a machine learning algorithm. [7] A computer-implemented method according to any one of the preceding claims, wherein the machine performs a response when an anomaly in the sensor data is detected. [8] A computer-implemented method according to any one of the preceding claims, wherein the sensor data is stored in a buffer and wherein the sensor data stored in the buffer is transferred to a protected memory when an anomaly has been detected. [9] A computer program comprising program code for carrying out a method according to any one of the preceding claims when the computer program is executed on a computer. [10] Computer-readable data carrier with program code of a computer program for carrying out a method according to one of claims 1 to 9 when the computer program is executed on a computer. [11] System for detecting anomalies in the use of a machine, wherein the system is designed to carry out a method according to one of claims 1 to 9, wherein the system comprises at least one computing unit and at least one smart sensor communicatively connected to the computing unit. [12] A machine comprising a system for detecting anomalies in the use of the machine according to claim 11.

Images