Multi-node system for synchronized data collection

Abstract

A method for operating a monitoring system (100) for a machine (110), the monitoring system comprising at least a first sensor unit (120) and a second sensor unit (120) attached at a respective first and second position on the machine (110). The first and second sensor units (120) are arranged to communicate sensor data over respective wireless links. The method comprises synchronizing (S1) the first sensor unit (120) and the second sensor unit (120) with respect to time, obtaining (S2) a first output from the first sensor unit (120) and a second output from the second sensor unit (120), the first and second output each comprising at least a measurement data relating to a movement of the machine (110) at the respective first and second position on the machine as a function of time, and determining (S3) a combined data set from the first output and the second output, wherein the combined data set relates a movement of the machine (110) at the first and second position as a function of time. The method further comprises extracting (S4) at least one operational parameter of the machine (110) from the combined data set and operating (S5) the monitoring system (100) of the machine (110) according to the at least one operational parameter.

Description

[0001] TITLE

[0002] MULTI-NODE SYSTEM FOR SYNCHRONIZED DATA COLLECTION

[0003] TECHNICAL FIELD

[0004] The present disclosure relates to systems and methods for monitoring the operation of machines, particularly heavy-duty machinery such as vibrating screens.

[0005] BACKGROUND

[0006] In industries that use heavy machinery, issues such as machine failure and / or excessive wear and tear on the machines lead to disturbances in production and increased costs relating to machine repair and replacement. Monitoring the operation of the machines is therefore crucial for identifying problems at an early stage and detecting disturbances and deviations in the operation of the machines. Such information can subsequently be used e.g. to adjust the operation of the machine and / or to plan maintenance so that it causes the least amount of disturbance.

[0007] Preferably, monitoring of heavy machinery is carried out in real time. Additionally, measurements taken at multiple points on each machine may be needed in order to provide sufficient data. This is especially the case for machines that move during operation and operate under heavy loads, such as vibrating screens.

[0008] WO2019006506 A1 discloses systems and methods for monitoring the operation of a vibrating screen by correlating measurements from multiple sensor nodes.

[0009] Still, there is a need for improved systems and methods for monitoring the operation of heavy machinery.

[0010] SUMMARY

[0011] It is an objective of the present disclosure to provide improved methods for monitoring heavy machinery. This objective is at least in part obtained by a method for operating a monitoring system of a machine. The monitoring system comprises at least a first sensor unit and a second sensor unit attached at a respective first and second position on the machine. The first sensor unit and the second sensor unit are arranged to communicate sensor data over respective wireless links. The method comprises synchronizing the first sensor unit and the second sensor unit with respect to time and obtaining a first output from the first sensor unit and a second output from the second sensor unit. The first and second output each comprise measurement data relating to a movement of the machine at the respective first and second position on the machine as a function of time. The method further comprises determining a combined data set from the first output and the second output, wherein the combined data set relates to movement of the machine at the first and second position as a function of time.

[0012] Moreover, the method comprises obtaining at least one operational parameter of the machine from the combined data set and operating the monitoring system of the machine according to the at least one operational parameter.

[0013] The method thus involves obtaining time-synchronized measurement data from at least a first and a second separate location on the machine. Due to the synchronization, data from the first and second location can be compared in the same time base, e.g. in order to identify differences in the movement of the machine at the two locations at a given point in time and also cause and effect like relations in the output from the different sensor units. This is an advantage compared to non- synchronous measurements which does not allow comparison of movement data as function of time at two or more different locations on the machine, i.e., time-analysis of the sensor output signals in a common time base. For instance, many types of heavy machinery exhibit motion which propagates over the machinery over time. An impulse or impact at one location on the machine is expected to cause changes in the output of the sensors arranged on the machine in dependence of the relative location of the sensors. By comparing the output of the sensors in the same time base it becomes possible to detect when something on the machine has changed.

[0014] Preferably, the measurement data comprised in the first and second output is raw sensor data. Raw sensor data is herein taken to mean data representing the output of the sensor without the application of any statistical operation such as the calculation of a mean value, a root mean square (RMS) value, or similar. Optionally, raw data may be down-sampled compared to the unfiltered output of the sensor in order to accommodate low bandwidths of a wired or wireless connection. Use of raw data in the combined data set allows for more flexibility when it comes to data analysis, as the analysis is not restricted by only having access e.g. to statistical properties of the measurement data. Due to the time synchronization of the sensor units, the raw data can be time aligned, e.g., at a main control unit in the monitoring system, so that the measurement data from the different locations on the machine can be analyzed as function of time in a common time base. For instance, if the raw data from the different sensors is placed in vectors, then the different vectors can be time shifted so as to be given in the same time base. An event recorded at one location can then be tracked over time as its consequences for the motion of the machine propagates to the other sensor locations. This time synchronized data provides more information compared to known sensor systems which only monitor, e.g., vibration magnitude as function of frequency without considering the time dependence between the different sensor output signals.

[0015] The monitoring methods described herein are particularly suitable for use with machine learning models which are often able to identify and exploit complex timerelationships between the different sensor signals. Such relationships are not possible to find and exploit if the data from the different sensor units is not time synchronized.

[0016] The combined data set can relate to a difference in the raw data acquired from sensor units placed at two different locations on the machine. This allows e.g. for identifying correlations between the movement at the first and second location, which is an advantage. A certain time correlation between the output signal of one sensor unit and the output signal of another sensor unit may be expected, in which case a decrease in this time correlation may be indicative of malfunction or discrepancy in the operation of the machine. Other malfunctions and error events may give rise to time correlations between the different sensor output signals that is not expected. Hence, by detecting a new unexpected time correlation certain error events can be detected.

[0017] According to some aspects, synchronizing the first sensor unit and the second sensor unit comprises transmitting a synchronization signal over a wireless link from a main control unit to the first sensor unit and the second sensor unit. Advantageously, this is a synchronization method that is easy to implement. However, other synchronization methods may also be used.

[0018] Determining the combined data set from the first output and from the second output may comprise calculating any of a cross-correlation function, a phase shift, and a relative time delay. These quantities represent features of the movement of the machine and enable the identification of correlations between the two or more sensor units.

[0019] Determining a combined data set from the first output and the second output may also comprise obtaining data relating to an expected movement of the machine as function of time and determining a difference between the expected movement of the machine and an actual movement of the machine as function of time. This is an advantage in machines such as vibrating screens which are arranged to move in a periodic pattern during operation. Identifying additional movement that is different from the expected movement can enable the identification of deviations and disturbances in the operation of the machine, which is an advantage.

[0020] According to some examples, the measurement data comprises any of an acceleration measurement, an angular rate measurement, and a magnetic field measurement. Such data can typically be obtained from an inertial measurement unit. Access to an acceleration measurement makes it possible to obtain also the position and velocity of the sensor unit, while the angular rate measurement enables characterization of rotational movements.

[0021] Any of the acceleration measurement, the angular rate measurement, and the magnetic field measurement may be performed in at least two dimensions. A two- or three-dimensional measurement provides information about the movement of the machine in all directions, which is an advantage.

[0022] The time synchronization of the sensor units described herein allows for use of less complex sensors. For instance, three or more one-dimensional accelerometers can be combined thanks to the common time base achieved by the synchronization into acceleration data in two or three dimensions. The same can be said for angular rate measurements and magnetic field measurements. It is an advantage that the monitoring systems described herein can be operated using less complicated one or two-dimensional sensor units.

[0023] The method may comprise training a machine learning model and / or a deep learning model to identify an operational condition in the machine using the measurement data and / or the combined data set. Such machine learning can then be used to analyse gathered data as will be described in more detail herein. According to some examples, the operational condition may be a fault condition. Furthermore, training the machine learning model may comprise using at least one set of time synchronized sensor data from at least one other machine. Using data from a different machine of the same or a similar type can increase the amount of available training data, which is an advantage. Training data from a different machine could also comprise e.g. data representing unusual operational conditions such as faults that occur relatively rarely. Using such training data can improve the performance quality of the machine learning model. Preferably, extracting at least one operational parameter of the machine from the combined data set comprises using a machine learning model to identify an operational condition of the machine from the combined data set. Identifying an operational condition may also comprise using a machine learning model to identify a fault condition of the machine.

[0024] The machine learning model may be a classical machine learning model such as a random forest or support vector machine, or it may comprise a deep neural network. Machine learning models can be applied either to raw data or to processed data such as cross-correlation functions and power spectra in order to identify patterns in the machine motion. Here, the machine learning model may for example be trained to detect patterns that correlate with errors or deviations in machine operation.

[0025] An advantage of using a machine learning model is that the model may identify patterns that indicate e.g. a machine error but that have not previously been identified using ordinary data analysis. Thus, use of machine learning models may enable a more comprehensive analysis of the available data, which is an advantage. The machine learning model may be able to exploit time dependencies between the different output signals of the sensor units which is not possible if the sensor units are not time synchronized in the manner described herein.

[0026] The method may also comprise transmitting measurement data and / or the combined data set to an external database, i.e., a database remote from the machine. A larger amount of data, either from multiple machines, over an extended period of time, or both, makes it easier to identify patterns and trends in the operation of the machines. As an example, new data obtained for a specific machine may be compared to earlier data from the same machine to identify changes in machine operation.

[0027] A computer program comprising program code means for performing the steps of the methods described above, when said program is run on a control system comprising one or more control units, is also described herein, as well as a computer readable medium carrying a computer program comprising program code means for performing the steps of the abovementioned methods, when said program product is run on a control system comprising one or more control units, and a computer program product comprising such a computer program, and a computer readable storage medium on which the computer program is stored.

[0028] The methods disclosed herein are associated with the same advantages as discussed above in connection to the different apparatuses. There is also disclosed herein computer programs, computer program products, and control units associated with the above-mentioned advantages.

[0029] Generally, all terms used in the claims are to be interpreted according to their ordinary meaning in the technical field, unless explicitly defined otherwise herein. All references to "a / an / the element, apparatus, component, means, step, etc." are to be interpreted openly as referring to at least one instance of the element, apparatus, component, means, step, etc., unless explicitly stated otherwise. The steps of any method disclosed herein do not have to be performed in the exact order disclosed, unless explicitly stated. Further features of, and advantages with, the present invention will become apparent when studying the appended claims and the following description. The skilled person realizes that different features of the present invention may be combined to create embodiments other than those described in the following, without departing from the scope of the present invention.

[0030] BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The present disclosure will now be described in more detail with reference to the appended drawings, where:

[0032] Figure 1 schematically illustrates a monitoring system with sensor units;

[0033] Figure 2 schematically illustrates an example sensor unit;

[0034] Figure 3 illustrates example sensor units mounted on a vibrating screen;

[0035] Figure 4 shows examples of measurement data from four sensor units;

[0036] Figure 5 is a flow chart illustrating methods;

[0037] Figure 6 schematically illustrates a control unit; and

[0038] Figure 7 schematically illustrates a computer program product.

[0039] DETAILED DESCRIPTION

[0040] Aspects of the present disclosure will now be described more fully with reference to the accompanying drawings. The different devices and methods disclosed herein can, however, be realized in many different forms and should not be construed as being limited to the aspects set forth herein. Like numbers in the drawings refer to like elements throughout. The terminology used herein is for describing aspects of the disclosure only and is not intended to limit the invention. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.

[0041] The following description focuses on methods for monitoring machinery in mining and ore processing, particularly vibrating screens. However, a person skilled in the art will realize that the systems and methods herein described can be applied also to other types of industrial equipment.

[0042] Vibrating screens are used to separate granulated materials, such as crushed ores, according to their particle size and density. A vibrating screen generally comprises at least one drive or motor that induces vibration, a screen medium arranged to separate particles, and a deck or frame on which the drive and screen medium are mounted. The screen medium is typically a mesh comprising a plurality of holes, where the size of the holes is selected in dependence of the size of the particles that the screen is intended to separate. The screen medium may be woven wire cloth, a punch plate, or be made from rubber or other polymer compounds. Many industrial screens comprise a plurality of screen media with different mesh sizes in order to be able to separate granulated materials into multiple grades.

[0043] Different types of vibrating screens are distinguished by factors such as how the vibration is generated and the direction of the vibration movement. The movement may for example be linear, circular, or elliptical. Vibrating screens can also be adapted to processing different types of materials, as well as to processing either dry or wet material. Generally, vibrating screens are known in the art.

[0044] Industrial processes involving heavy machinery such as vibrating screens necessarily have to be stopped if the machinery breaks down, which can lead to costly interruptions in operation. Some processing facilities are automated, or semiautomated, and in such cases the interruption in operation becomes more severe. For example, in case surrounding machinery such as transport belts and feed systems are not stopped when another machine such as a vibrating screen suffers malfunction, this can lead to additional damage to the equipment.

[0045] Improved monitoring of the operation of machines such as vibrating screens can make it possible to detect problems in advance and plan maintenance so that the risk of breakdowns and unplanned interruption is minimized. Additionally, information about the operation of the machinery can be used to improve performance. For example, information about movement patterns of a vibrating screen can be used for automatic adjustments to the machine in order to optimize machine operation.

[0046] Figure 1 shows a monitoring system 100 arranged to monitor a machine 110. The monitoring system shown in the figure comprises four sensor units 120 in communication with a main control unit 130. The communication takes place over wireless link. Wireless communication is generally known and will therefore not be discussed in more detail herein. A monitoring system 100 according to the teachings herein could also comprise only two or three sensor units 120, or a larger number of sensor units such as six or eight sensor units. The sensor units can be of the same type or of different type, i.e., configured to output the same type of data or different types of data.

[0047] The main control unit 130 is arranged to communicate with the sensor units 120 in the system 100, e.g. in order to receive measurement data from the sensor units 120. The communication takes place over a wireless connection with the main control unit 130 arranged to send and receive signals using electromagnetic waves, particularly in the radio frequency spectrum. The wireless connection may use e.g. Wi-Fi, 4G, or 5G. It may also use a radio protocol such as Bluetooth Low Energy (BLE), or other wireless communication means.

[0048] Figure 1 shows a system comprising a main control unit 130 that is housed in a single device. However, the main control unit 130 may also be distributed over a range of devices. In particular, if the main control unit 130 communicates with the sensor units 120 over a wireless connection, multiple transceivers may be used in order to ensure a reliable communication channel to each sensor unit 120. The number and placement of transceivers may for example be selected depending on the environment around the machine 110, the shape and size of the machine 110, and the number of sensor units 120 in the system.

[0049] The main control unit 130 may also be arranged to send data to one or more external units 140, and to receive data from the one or more external units 140. This may be accomplished over a wired connection such as an ethernet or CAN connection, or over a wireless connection using e.g. Wi-Fi, 4G, or 5G. The external devices 140 may comprise a data storage device, a local or online server, a mobile device such as a smartphone, or any other suitable device.

[0050] Each sensor unit 120 is arranged to collect measurement data relating to the operation of the machine 110. The type of measurement data being collected depends on the relevant operational parameters of the machine 110. If the machine 110 is a machine that moves during operation, such as a vibrating screen, the measurement data may comprise data representing any of an acceleration, a velocity, an angular rate and a position. However, the measurement data may also comprise measurements of any of an ambient temperature, an ambient pressure, a humidity, a magnetic field, or any other relevant property. The measurement data may furthermore comprise image data and audio obtained by one or more vision sensors and microphones, respectively.

[0051] Figure 2 schematically illustrates a sensor unit 120 comprising a sensor 121 , an energy providing device 122, and a control unit 123. In the figure, the sensor is represented as a tri-axial accelerometer. However, the sensor 121 could also be any of a gyroscope, a temperature sensor, a pressure sensor, a humidity sensor, a magnetometer, or any other suitable sensor. It is also possible that the sensor unit 120 comprises sensors of more than one type, such as an accelerometer to detect translational movement and a gyroscope to measure rotation.

[0052] The energy providing device 122 is a device that is arranged to supply the sensor unit with energy. Preferably, the energy providing device 122 does not require the sensor unit 120 to be connected to an external energy source via a cable. The energy providing device can thus be an energy storage device such as a battery or a supercapacitor. The energy providing device 122 preferably comprises an energy harvesting device arranged to harvest kinetic energy, e.g. vibrational energy.

[0053] The sensor unit 120 also comprises a control unit 123 arranged to control operation of the sensors, and communication means arranged to communicate wirelessly with the main control unit 130.

[0054] The sensor units 120 are attached to the machine 1 10 using attachment means. The attachment means may for example be a magnetic mount, a bracket, a quick-release plate, or any other suitable attachment means. The location of each sensor unit on the machine 1 10 can be chosen depending on the type of machine 1 10 and on the type of measurement data that is to be gathered. According to one example, sensor units 120 can be distributed evenly over the body of the machine to gather data from a wide range of positions. According to another example, sensor units can be placed at points of particular interest, such as close to a drive shaft. Figure 3 shows four sensor units 120-1 , 120-2, 120-3 and 120-4 mounted at the corners of a vibrating screen 1 10. With reference to Figure 5, and also Figure 1 , there is disclosed a method for operating a monitoring system 100 for a machine 1 10. The monitoring system comprises at least a first sensor unit 120 and a second sensor unit 120 attached at a respective first and second position on the machine 110. The first sensor unit 120 and the second sensor unit 120 are arranged to communicate sensor data over wireless links. The method comprises synchronizing S1 the first sensor unit 120 and the second sensor unit 120 with respect to time and obtaining S2 a first output from the first sensor unit 120 and a second output from the second sensor unit 120. Here, the first and second output each comprise measurement data relating to a movement of the machine 1 10 at the respective first and second position on the machine as a function of time. The method also comprises determining S3 a combined data set from the first output and the second output, wherein the combined data set relates a movement of the machine 110 at the first and second position as a function of time.

[0055] Furthermore, the method comprises obtaining S4 at least one operational parameter of the machine 1 10 from the combined data set, and operating S5 the monitoring system 100 of the machine 1 10 according to the at least one operational parameter.

[0056] For the first and second sensor unit 120 to be synchronized is herein taken to mean that for each time t at which measurement data from both the first and second sensor unit 120 is available, the measurement data from the first sensor at time t and the measurement data from the second sensor at the same time t can be identified and compared. This can be accomplished in several different ways.

[0057] According to some aspects, measurement data can be accompanied by a timestamp. The timestamp may be based on a global reference time, which can for example be obtained from a satellite navigation system such as GPS or Galileo. The global reference time can also be obtained from one of the control units comprised in the system 100, such as the main control unit 130 or a control unit 123 comprised in a sensor unit 120.

[0058] A timestamp can also be generated using a local reference time, such as the internal time of the control unit 123 in the sensor unit 120. In this case, the synchronization step may comprise determining a discrepancy between the internal time of the control unit 123 of the first sensor unit 120 and the internal time of the control unit 123 of the second sensor unit 120, and / or determining a response time between the main control unit 130 and each sensor unit 120. According to other aspects, synchronization with respect to time can be accomplished by determining a time offset between data streams from the first and second sensor unit 120 at the main control unit 130. The relative time offset can for example be found by transmitting test messages from the first and second sensor units 120 to the main control unit 130 and registering a difference in time of arrival. Alternatively, an absolute delay can be determined for each sensor unit. The absolute delay can represent e.g. the time interval between the sensor registering a measurement and the arrival of the corresponding data at the main control unit 130. As such, it can depend on the properties of the sensor and on the wireless link between the sensor unit 120 and the main control unit 130, among other factors.

[0059] There are several known methods for performing synchronization of wireless sensor nodes or sensor units. Such methods include but are not limited to sender-receiver synchronization methods, sender-sender synchronization methods, and reference broadcast synchronization. These methods are known in the art and will therefore not be described in detail herein.

[0060] According to one example, the main control unit 130 is arranged to transmit a first synchronization signal to at least one sensor unit 120. Optionally, the at least one sensor is arranged to transmit a request for synchronization to the main control unit. The main control unit can be arranged to receive the request for synchronization and to capture a timestamp. The main control unit can subsequently transmit the captured timestamp to the sensor.

[0061] Preferably, the time synchronization should be performed with sufficient accuracy and precision such that the relative motion of the sensor units can be resolved. As an example, if the machine is performing a periodic motion, the time synchronization should be sufficiently accurate such that it is possible to tell from the data if two or more sensor units are moving in or out of phase. According to an example, a deviation in the time synchronization between the main control unit and a sensor unit should be no more than 1 ms. Furthermore, the accuracy of the time synchronization is preferably such that any time synchronization error is less than 1 ms, or such that the synchronization tolerance is at most 1 ms.

[0062] Put another way, synchronizing S1 the first sensor unit 120 and the second sensor unit 120 may comprise transmitting S11 a synchronization signal from a main control unit 130 to the first sensor unit 120 and the second sensor unit 120. According to another example, one sensor unit 120 is configured as a reference sensor unit. The internal clocks of all other sensor units 120 are subsequently synchronized to a reference time provided by the internal clock of the reference sensor unit. This is accomplished over a wireless connection, for example using a wireless radio protocol such as Bluetooth Low Energy (BLE). The reference time is preferably provided with a resolution of 1 ms or less.

[0063] It may be noted that any synchronization method described herein can be generalized from a system comprising only two sensor units 120 to a system comprising four, six or eight sensor units, or any number of sensor units suitable for the machine 1 10 to be monitored.

[0064] As previously described, a sensor unit 120 preferably comprises an accelerometer, but may also comprise other sensor such as rotation sensors, magnetic field sensors, temperature sensors, humidity sensors, etc. The measurement data may therefore comprise any of an acceleration measurement, an angular rate measurement, and a magnetic field measurement. It may also comprise velocity data and / or position data and, depending on the synchronization method used, a timestamp. However, the output may also comprise temperature data, humidity data, pressure data, or any other relevant measurement data.

[0065] Due to the time synchronization implemented by the monitoring systems described herein, it is possible to use lower dimensional sensor types and post-process the sensor data into higher dimensional data. For instance, several one-dimensional accelerometers can be deployed over the machine. Each one-dimensional accelerometer sensor provides a data stream of one-dimensional accelerometer data. By time-aligning the different data streams, which is enabled by the step of synchronization, the main control unit 130 can obtain multi-dimensional accelerometer data.

[0066] Alternatively, it is possible to use multidimensional sensors that measure e.g. acceleration or angular rate in two or three dimensions. Accordingly, the method may comprise performing one or more of the acceleration measurement, the angular rate measurement, and the magnetic field measurement in at least two dimensions.

[0067] Preferably, the measurement data comprised in the first and second output is raw sensor data. Raw data is herein taken to mean data representing the output of the sensor without the application of any statistical operation such as the calculation of a mean value, a root mean square (RMS) value, or similar. Optionally, raw data may be down-sampled compared to the unfiltered output of the sensor in order to accommodate low bandwidths of a wired or wireless connection. Use of raw data in the combined data set allows for more flexibility when it comes to data analysis, as the analysis is not restricted by only having access e.g. to statistical properties of the measurement data.

[0068] According to some alternatives, determining S3 a combined data set may comprise combining measurements from the first output and the second output into an aggregated data set.

[0069] Figure 4 illustrates an aggregated data set comprising data from four sensor units 120 plotted against time. The sensor units are labelled 120-1 , 120-2, 120-3, and 120-4, as in Figure 3, with the measurement data from each unit labelled as Data-1 , Data-2, Data-3, and Data-4 respectively. The plots in Figure 4 show Data-1 as comprising a peak at a time t0. Data-2 comprises a similar peak at t1 ;Data-3 at t2, and Data-4 at t3. As the sensor units 120 are time synchronized, it is possible to line up the data from each sensor unit with the others in time and determine in what order the sensor units 120 register the peak as well as the duration between the pulse being registered e.g. by the first sensor unit 120-1 and any of the other sensor units. For example, if the data is accelerometer data and the peak represents a disturbance in the movement of the machine 1 10, it would be possible to see in which part of the machine the disturbance originates and how quickly it travels along the machine 1 10. The aggregated data set thus relates to a difference in the movement of the machine 1 10 at the positions of each of the sensor units 120.

[0070] Consider a vibrating screen where a boom or other structural element has broken. This structural element will no longer transmit vibration, and consequently an expected movement at some sensors will no longer be registered, indicating the malfunction.

[0071] In case two sensor units on a machine are mechanically separated from each by a damping element, and an increase in the time correlation in the output signals from the two sensor units is detected, then it is likely that the damping element has suffered malfunction.

[0072] Determining S3 a combined data set may include combining the data so as to obtain information about a difference in the movement of the machine at the locations where the sensor units 120 are attached, e.g. the first and second locations mentioned above. For example, determining S3 a combined data set may comprise calculating a difference between measurement data in the first output and the measurement data in the second output. Optionally, determining S3 a combined data set from the first output and the second output comprises calculating S31 any of a cross-correlation function, a phase shift, and a time delay.

[0073] Calculating the cross-correlation function may herein entail selecting a particular type of measurement data, such as accelerometer data, from two sensor units 120 and obtaining the measurement data in the form of time series. Subsequently, the crosscorrelation between these time series of measurement data can be calculated.

[0074] Referring again to Figure 4, a cross-correlation function could for example be calculated between Data-1 and Data-2. This cross-correlation function would have a peak at a point equal to the difference in time between, Likewise, the crosscorrelation between Data-1 and Data-3 would show a peak at T = t2- t0.

[0075] A phase shift can be relevant if the machine 1 10 is in periodic motion during operation, as is the case e.g. for vibrating screens. Each sensor unit 120 will then also be in periodic motion. A relative phase shift between two sensor units 120 can then be calculated based on where in the periodic motion each sensor unit 120 is at a time tola a system 100 comprising more than two sensor units 120, the combined data set is preferably a combined data set obtained from the output of all sensor units 120 in the system. In cases where the combined data set comprises a difference between measurement data from different sensor units 120, one sensor unit 120 may be configured as a reference sensor unit. A difference may then be calculated between measurement data from each remaining sensor unit 120 and measurement data from the reference sensor unit. The same reasoning applies to the calculation of other metrics such as cross-correlation functions and phase shifts.

[0076] Some machines 110 are arranged to move according to a predetermined pattern when in operation. This is for example the case with vibrating screens. In such cases, determining S3 a combined data set from the first output and the second output may comprise obtaining S32 data relating to an expected movement of the machine 1 10 as a function of time and determining a difference between the expected movement of the machine and an actual movement of the machine as a function of time.

[0077] The expected movement of the machine may be a predetermined movement characterized by parameters such as a direction, a frequency or period, and an amplitude. More complex movement patterns where the machine moves in different directions to different degrees are also possible. This predetermined movement may be made available to the system 100 e.g. as user input. Data relating to additional movement could then be identified as data indicating deviations from the expected movement.

[0078] Alternatively, the system 100 may be arranged to identify the expected movement and the additional movement from the combined data set. Such identification may be carried out using a machine learning model and / or a deep learning model.

[0079] Determining S3 a combined data set can be performed by any control unit comprised in the system 100. As an example, the determination of the combined data set is performed by a main control unit 130 or an external unit 140.

[0080] The method comprises obtaining S4 at least one operational parameter of the machine 110 from the combined data set. An operational parameter is herein taken to be any parameter that describes or quantifies the operation of the machine, such as a frequency or an amplitude. An operational parameter may also be more complex, such as a periodic pattern or movement or an indicator for a fault condition.

[0081] The method also comprises operating S5 the monitoring system 100 of the machine 1 10 according to the at least one operational parameter. According to one example, if the operational parameter is an indicator for a fault condition, operating S5 the monitoring system 100 of the machine 1 10 according to the operational parameter may comprise sending an alert to an operator of the machine 1 10 indicating the fault. This may also be the case if simpler operational parameters such as a frequency or amplitude fall outside a predetermined acceptable interval. Alternatively, the system 100 may simply output the operational parameter to a device such as a display where it can be observed by a system operator.

[0082] According to some alternatives, the monitoring system 100 is connected to a control unit that controls at least in part the operation of the machine 110. Operating S5 the monitoring system 100 may then comprise automated adjustments of the operation of the machine in dependence of one or more operational parameters obtained from the combined data set. The adjustments can for example serve the purpose of preventing fault conditions from occurring or of optimizing machine operation.

[0083] Preferably, obtaining S4 at least one operational parameter of the machine 110 from the combined data set comprises using S41 a machine learning model to identify an operational condition of the machine from the combined data set. According to some alternatives, the machine learning model can be used to identify S411 a fault condition of the machine.

[0084] Here, a machine learning model can refer both to classical machine learning models and to deep learning models comprising neural networks. The machine learning model can be a model trained to identify errors and deviations in the operation of the machine based on the first and second output and / or processed data such as correlation functions and power spectra.

[0085] As one example, the machine learning model may be a classical machine learning model employing methods such as regression analysis, clustering algorithms, or classification algorithms. In particular, clustering algorithms such as k-means, hierarchical clustering, and density-based spatial clustering with noise (DBSCAN) can be used to group similar motion sequences, categorize different types of movements, and classify recurring motion patterns.

[0086] As another example, the machine learning model may be a deep learning model comprising a neural network such as a recurrent neural network (RNN), convolutional neural network (CNN), autoencoders, and / or transformer networks. Among these examples, RNNs such as long short-term memory (LSTM) networks are particularly well suited for time-series analysis due to their ability to maintain memory of previous states. LSTM networks can learn long-term dependencies in motion data and predict future movements based on past sequences.

[0087] CNNs function through the consecutive application of a plurality of convolutional filters to the data, which results in the detection of local patterns and trends. They can be adapted for motion data analysis, e.g. by treating time-series data as a onedimensional signal.

[0088] Autoencoders can be used for unsupervised learning and are frequently employed for anomaly detection. By encoding the motion data into a lower-dimensional space and subsequently reconstructing it, autoencoders can highlight key features and anomalies in the motion patterns.

[0089] Transformer architectures comprise self-attention mechanisms that enable the modelling of dependencies across different time steps. They thus provide a powerful tool for capturing complex motion dynamics. The method may also comprise transmitting S6 measurement data and / or the combined data set to a database. The external unit 140 shown in Figure 1 may for example represent such a database. The database may be used to store measurement data from one machine 1 10, or from multiple machines. Data collected from one or more machines may be used to train a machine learning model as discussed below.

[0090] According to some aspects, the method comprises training SO a machine learning model and / or a deep learning model to identify an operational condition in the machine 110 using the measurement data and / or the combined data set. The operational condition may be a fault condition.

[0091] As mentioned above, the machine learning model may be a classical machine learning model employing methods such as regression analysis, clustering algorithms, or classification algorithms. The process of training a classical machine learning model comprises steps such as data collection and preprocessing, feature engineering, model selection, and model training.

[0092] In the present case, data collection consists of selecting data gathered from a system 100 monitoring a machine 100, as shown in Figure 1 . The data may e.g. be stored in a database as described above. Preprocessing of the data may comprise data cleaning steps such as removal of outliers, or it may comprise feature normalization or other data processing steps that are suited to the selected machine learning model. Feature engineering entails extracting relevant features from the raw data, for example statistical measures such as a mean or variance and frequency domain features such as power spectra. Feature engineering may also comprise calculating e.g. a phase shift or cross-correlation function.

[0093] Subsequently, the data is divided into a training data set and a validation data set. The training data set is used to train the chosen model, e.g. by automatic adjustment of the model parameters to minimize an error in the model output. The model is subsequently validated using the validation data.

[0094] Optionally, the model may be periodically re-trained after deployment. This makes it possible to adapt the model to changes in e.g. machine operation.

[0095] As has also been mentioned above, the machine learning model may be a deep learning model comprising a neural network such as a recurrent neural network (RNN), convolutional neural network (CNN), autoencoders, and / or transformer networks.

[0096] T raining of a deep learning model shares some steps with the training of conventional machine learning models as described above, such as the data collection and preprocessing steps. However, feature engineering is less crucial as deep learning models are typically arranged to learn features automatically. Model training methods also differ as loss functions and optimization algorithms may need to be selected to suit the selected type of deep learning model and the problem at hand. Furthermore, learning strategies such as unsupervised learning and reinforcement learning may be applied, e.g. when using models such as autoencoders which can detect anomalies in data by determining a model reconstruction error and checking if it exceeds a predetermined threshold. Anomaly detection using autoencoders is known in the art.

[0097] As with conventional machine learning models, deep learning models can advantageously be re-trained using new data if needed.

[0098] A machine learning model used to identify an operational condition of one machine can be trained using not only data from that machine, but also from other machines that may be e.g. of a similar type or of the same model. According to some examples, training SO the machine learning model to identify an operational condition in the machine 1 10 thus comprises using at least one set of time synchronized sensor data from at least one other machine. A set of sensor data can here refer to a sequence of data from one or more sensor types as a function of time. As an example, the set of sensor data may be sensor data from two or more accelerometers.

[0099] There is also herein disclosed a computer program comprising program code means 720 for performing the steps of the methods discussed above, when said program is run on a control system comprising one or more control units. Furthermore, there is herein disclosed a computer readable medium 710 carrying a computer program comprising program code means 720 for performing the steps of the abovementioned methods, when said program product is run on a control system comprising one or more control units. Also, there is disclosed a computer program product 700 comprising such a computer program 720, and a computer readable storage medium 710 on which the computer program is stored.

[0100] Figure 6 schematically illustrates, in terms of a number of functional units, the general components of the control unit 600, 130, 123. Processing circuitry 610 is provided using any combination of one or more of a suitable central processing unit CPU, multiprocessor, microcontroller, digital signal processor DSP, etc., capable of executing software instructions stored in a computer program product, e.g. in the form of a storage medium 630. The processing circuitry 610 may further be provided as at least one application specific integrated circuit ASIC, or field programmable gate array FPGA.

[0101] Particularly, the processing circuitry 610 is configured to cause the monitoring system 100 to perform a set of operations, or steps, such as the methods discussed in connection to Figure 5, and the discussions above, and also to set operating parameters of the system according to the discussions above. For example, the storage medium 630 may store the set of operations, and the processing circuitry 610 may be configured to retrieve the set of operations from the storage medium 630 to cause the device to perform the set of operations. The set of operations may be provided as a set of executable instructions. Thus, the processing circuitry 610 is thereby arranged to execute methods as herein disclosed.

[0102] The storage medium 630 may also comprise persistent storage, which, for example, can be any single one or combination of magnetic memory, optical memory, solid state memory or even remotely mounted memory. This storage medium may be configured to store one or more sets of configuration settings for the monitoring system 100.

[0103] The control unit 130 may further comprise an interface 620 for communications with at least one external device. As such the interface 620 may comprise one or more transmitters 601 and receivers 602, comprising analogue and digital components and a suitable number of ports for wireline or wireless communication.

[0104] The processing circuitry 610 controls the general operation of the control unit 600, 130, 123 e.g., by sending data and control signals to the interface 620 and the storage medium 630, by receiving data and reports from the interface 620, and by retrieving data and instructions from the storage medium 630.

[0105] Figure 7 illustrates a computer readable medium 710 carrying a computer program comprising program code means 720 for performing the methods illustrated in Figure 5 and / or for executing the various functions discussed above, when said program product is run on a computer. The computer readable medium and the code means may together form a computer program product 700. This computer program product may comprise one or more sets of configurations for controlling the device 100 discussed above to perform the methods disclosed herein.

Claims

CLAIMS1. A method for operating a monitoring system (100) of a machine (1 10), the monitoring system comprising at least a first sensor unit (120) and a second sensor unit (120) attached at a respective first and second position on the machine (1 10), the monitoring system further comprising at least one control unit (130, 123), where the first sensor unit (120) and the second sensor unit (120) are arranged to communicate sensor data over respective wireless links, the method comprising: synchronizing (S1 ) the first sensor unit (120) and the second sensor unit (120) with respect to time, obtaining (S2) a first output from the first sensor unit (120) and a second output from the second sensor unit (120), the first and second output each comprising measurement data relating to a movement of the machine (1 10) at the respective first and second position on the machine as a function of time, determining (S3) a combined data set from the first output and the second output, wherein the combined data set relates to movement of the machine (1 10) at the first and second position as function of time, obtaining (S4) at least one operational parameter of the machine (1 10) from the combined data set, and operating (S5) the monitoring system (100) of the machine (1 10) according to the at least one operational parameter.

2. The method according to claim 1 , wherein synchronizing (S1 ) the first sensor unit (120) and the second sensor unit (120) with respect to time comprises transmitting (S11 ) a synchronization signal over a wireless link from a main control unit (130) to the first sensor unit (120) and to the second sensor unit (120).

3. The method according to claim 1 or 2, wherein determining (S3) the combined data set from the first output and the second output comprises calculating (S31 ) any of a cross-correlation function, a phase shift, and a relative time delay.

4. The method according to any previous claim, wherein determining (S3) the combined data set from the first output and from the second output comprises obtaining (S32) data relating to an expected movement of the machine (1 10) as a function of time and determining a difference between the expected movement of the machine and an actual movement of the machine as function of time.

5. The method according to any previous claim, wherein the measurement data comprises any of an acceleration measurement, an angular rate measurement, and a magnetic field measurement.

6. The method according to claim 5, comprising performing any of the acceleration measurement, the angular rate measurement, and the magnetic field measurement in at least two dimensions.

7. The method according to any previous claim, comprising training (SO) a machine learning model to identify an operational condition in the machine (110) using the combined data set.

8. The method according to claim 7, wherein the operational condition is a fault condition.

9. The method according to claim 7 or 8, wherein training (SO) the machine learning model to identify an operational condition in the machine (1 10) comprises using at least one set of time synchronized sensor data from at least one other machine.

10. The method according to any previous claim, wherein extracting (S4) at least one operational parameter of the machine (110) from the combined data set comprises using (S41 ) a machine learning model to identify an operational condition of the machine from the combined data set.1 1. The method according to claim 10, comprising identifying (S41 1 ) a fault condition of the machine using a machine learning model.

12. A computer program comprising program code means (720) for performing the steps of any of claims 1 to 1 1 , when said program is run on a control system comprising one or more control units.

13. A computer readable medium (700) carrying a computer program comprising program code means (720) for performing the steps of any of claims 1 to 11 , when said program product is run on a control system comprising one or more control units.

14. A computer program product (700) comprising a computer program (720) according to claim 12, and a computer readable storage medium (710) on which the computer program is stored.

Images