Monitoring method for application plants and said application plants

Abstract

The present invention relates to a method for monitoring an application plant for applying a coating agent, particularly a painting plant for painting automotive body components, comprising the following steps: - Determination of the first raw sensor data (16) representing the operating variables of the application plant, - Acquisition of a first control signal for controlling an application plant, and - Features (15) are extracted from the first raw sensor data (16), and the features (15) include a reduced amount of data compared to the first raw sensor data (16) and serve as a data basis for a machine learning algorithm. In addition, the present invention includes an application plant that is monitored accordingly.

Description

Technology Field

[0001] The present invention relates to a method for monitoring an application plant for applying an application agent, particularly a painting plant for painting automotive body components. Additionally, the present invention includes an application plant designed to correspond. Background Technology

[0002] In modern painting plants for painting automotive body components, numerous sensors detect operating variables and generate raw sensor data, and since operations are controlled by control signals, the operating status of the painting plant can be monitored by evaluating the raw sensor data and control signals. However, the monitoring methods used to monitor the operation of painting plants to date are not yet sufficiently satisfactory. The problem to be solved

[0003] DE 10 2007 062 132 A1 discloses a monitoring method for an application plant for applying a coating agent, particularly for a painting plant for painting automotive body components. Herein, raw sensor data is determined, and a control signal is recorded and used for function check.

[0004] DE 10 2018 214 170 A1 discloses a method and apparatus for performing a quality evaluation of a technical system using raw sensor data ("image data") and a control signal ("observation condition data") that serve as the data base for a machine learning algorithm. means of solving the problem

[0005] Publication DE 10 2016 217 948 A1 describes a method for predicting the quality of an adhesive joint where adhesive is applied by an adhesive processing device having an adhesive application device. Analysis of the measured data is performed by comparison with a data model. The statistical data model generated by the multivariate analysis method can be optimized by machine learning.

[0006] Accordingly, the present invention is based on the task of creating an improved monitoring method and a corresponding application plant.

[0007] This task is solved by the monitoring method according to the present invention or by the corresponding application plant according to the independent claims. Effects of the invention

[0008] The present invention is based on the technical knowledge that machine learning can be advantageously used to monitor application plants, as known from Fayyad U.; Patetsky-Shapiro, G.; Smyth, P.: "From Data Mining to Knowledge Discovery in Databases", AI Magazine, Vol. 17, No. 3 (1996).

[0009] Another literature source on machine learning methods is Bremer, M. "Principles of Data Mining", Springer-Verlag (2009).

[0010] According to the disclosed monitoring method, the monitoring method according to the present invention first allows determining first raw sensor data that reflects operating variables of an application plant. For example, the first raw sensor data may reflect the molding air pressure of the molding air used to form the spray jet of a rotary atomizer as an operating variable. However, the first raw sensor data may also reflect other operating variables of the application plant, as described in more detail below.

[0011] In addition, the monitoring method according to the present invention also stipulates that a first control signal is acquired to control an application plant in accordance with known monitoring methods. For example, the first control signal may define a target value of the molding air flow used to form a spray of a rotary atomizer. However, the first control signal may also be used to set other operating variables of the application plant, as described in detail below.

[0012] Accordingly, control signals together with raw sensor data form signal sets evaluated within the scope of the present invention. For example, the signal set may consider a setpoint value of the molding air flow as a control signal and an actual signal of the molding air pressure as raw sensor data. However, the signal set may include one or more actual signals and one or more control signals. Generally, the signal set may include 1...m different control signals and 1...k different actual signals, thereby the signals may also belong to different control loops.

[0013] The monitoring method according to the present invention is characterized by extracting so-called features from first raw sensor data, including a reduced amount of data compared to the first raw sensor data, and serving as a data basis for a machine learning algorithm.

[0014] The term "feature" as used in the context of the present invention first means that the amount of data is reduced by feature extraction, that is, the extracted features have a reduced amount of data compared to the raw sensor data.

[0015] Furthermore, the concept of "feature" as used in the context of the present invention implies that the feature is derived from an accurate operation process without losing information related to state detection. For example, it is no longer necessary to compare only exactly identical sequences of identical workpiece variations (e.g., car body variations). Rather, it is sufficient to compare, for example, similar "brushes" within completely different car bodies.

[0016] Furthermore, the term "feature" as used in the context of the present invention implies that the features are suitable as a data basis for machine learning algorithms, and that machine learning algorithms known by themselves are based on the extracted features. Thus, the features form a bridge between raw sensor data on the one hand and machine learning algorithms on the other, which enables stable data evaluation by machine learning algorithms that can often be mastered with limited resources.

[0017] In a preferred embodiment of the present invention, first and second raw sensor data reflecting different operating variables of an application plant are determined, and features are extracted from both the first and second raw sensor data to function as a database for a machine-learning algorithm. For example, the first raw sensor data may represent the molding airflow as an operating variable, while the second raw sensor data may represent the speed of the drive turbine of a rotary atomizer. However, the first and second raw sensor data may also reflect other operating variables of the application plant, as described in more detail below.

[0018] Preferably, feature extraction from the first and second raw sensor data is performed as follows.

[0019] In one step, an observation time window is first defined in which the first or second raw sensor data can be evaluated. For example, the observation time window may be a car body cycle, that is, the period between the supply of a motor vehicle body to be painted in a paint booth and the supply of the next motor vehicle body into the paint booth. This has the advantage of enabling status and monitoring during the normal production process. However, regarding the viewing window, there are various possibilities within the scope of the present invention, which will be described in detail below.

[0020] In another step, the observation time window is divided into individual sections and so-called comparison periods. In this way, comparison periods can be defined by changes (jumps) in control signals. For example, a new comparison period begins whenever a control signal related to the application changes within the body cycle. For example, the set point of the molding air flow of a rotary atomizer can be evaluated for this purpose. However, within the scope of the present invention, various other control signals may also be evaluated to determine the comparison period.

[0021] In the additional step, the comparison period itself is subdivided into several consecutive subsections, and the subdivision is predetermined by parameters.

[0022] For example, the comparison period can be divided into subsections as 0 ... n initial regulation steps, 0 ... m subsequent regulation steps, and subsequent residual steps as 0 ... 1 subsection.

[0023] The control steps are in the initial time step within the comparison time period, and at this time, the first and / or second raw sensor data responds to the jump of the first and / or second control signal so that the stop state cannot yet be said.

[0024] The regulated phases are periods after the regulated phase in which the first and / or second raw sensor data responds at least partially to the control signal jump.

[0025] To avoid error, it should be noted that the method itself does not check whether raw sensor data is emitted during the adjusting steps or the adjusted steps. Rather, the subdivision into the adjusting steps and the adjusted steps is fixed by parameters, namely, by the predefined duration of the adjusting steps and the predefined duration of the adjusted steps.

[0026] On the other hand, the residual step is the remaining period from the control steps and the controlled steps until the comparison period ends. In this case, the comparison period may include a single residual step or may not include any residual steps at all. For example, if the sum of the controlling steps and the controlled steps is exactly the same as the duration of the comparison period, the comparison period does not include any residual steps.

[0027] It should be noted that the regulating and residual steps are optional when the comparison period is sufficiently long to include not only the regulating steps but also the regulating steps and residual steps. However, if the comparison period is relatively short, the comparison period may be divided only into the regulating steps and residual steps. If the comparison period is very short, the comparison period may include only one regulating step and may not include any regulating steps at all.

[0028] In an additional step, at least one statistical parameter is calculated for individual subsections for the first and / or second raw sensor data, and if necessary, for additional raw sensor data within the individual subsections, and the calculated statistical parameter becomes an extracted feature.

[0029] It was briefly mentioned above that the comparison period can be limited by a jump or change in the control signal; that is, one change in the control signal starts the comparison period, and the next change in the control signal ends the current comparison period and starts the next comparison period.

[0030] In one variant of the present invention, the comparison period is limited by a change (jump) of the same control signal. In this case, when determining the comparison period, for example, the set point for the molding air flow of a rotary atomizer, only one control signal is evaluated. On the other hand, possible jumps of other control signals are not used to determine the comparison period in this variant of the present invention.

[0031] On the other hand, in another variation of the invention, multiple control signals are used to define a comparison period. Here, too, the comparison period is limited by changes or jumps in the control signals. However, different types of control signals are evaluated; for example, a jump in one control signal may indicate the start of a comparison period, while a jump in another control signal may indicate the end of the current comparison period and the start of the next comparison period.

[0032] It has already been mentioned that the statistical characteristics determined in individual subsections of the comparison period are components of the extracted features. However, it is also possible to include additional information regarding the extracted features within the scope of the present invention. For example, the features may include, for the sake of example, control signals after the last change and the amount of change of the set point over the time period.

[0033] In addition, it has already been mentioned above that the comparison period is subdivided into subsections, namely 0...n controlling steps, 0...m controlled steps, and 0...1 residual steps. At this point, it is noted again that the comparison period does not necessarily have to include several controlling steps and several controlled steps. For example, it is possible for the comparison period to include only one controlling step. The number of subsections and the durations of the subsections can be determined here as a function of the time constant of the raw sensor data. For this purpose, the time constant of the first and / or second raw sensor data is determined, and the number of subsections of the comparison period and the duration of one controlling step and / or one controlled step are determined as a function of the determined time constant. Only one time constant is selected per use case, which may also include multiple control and sensor data.

[0034] However, the duration of the regulating phase and / or individual regulated phases may be determined randomly.

[0035] It has already been mentioned above that the observation time window can be, for example, a body cycle. This is a component-related period associated with the component to be coated, that is, specifically, the car body component to be painted. However, otherwise, the observation time window can be another component-related period, such as the time required to prepare the application plant for the subsequent coating of the component to be coated.

[0036] Additionally, within the scope of the present invention, the observation time window may be an application plant-related time period, such as a time period of a setting operation, a test operation, a manual operation, or a maintenance operation.

[0037] In addition, the observation time window can be a time-related period such as one hour, one day, one week, one month, one quarter, or one year. Another example of a time-related period as an observation time window is the shift length of a task shift.

[0038] It was already mentioned above that statistical parameters are determined by individual subsections, which are components of the extracted features.

[0039] In one variant of the invention, this statistical parameter is a single-variable parameter that considers only one type of raw sensor data. Examples of such a single-variable statistical parameter include an arithmetic or geometric mean, a median, a variance, and a maximum or minimum value within a subsection, wherein the single-variable statistical parameter is always associated with the same raw sensor data, namely, a single operating variable.

[0040] In another variation of the present invention, the statistical parameter is a multivariate statistical parameter calculated from different raw sensor data. Examples of such multivariate statistical parameters include Pearson correlation coefficients and rank correlation coefficients. However, the multivariate statistical parameter may also consider three or more different raw sensor data of a signal set.

[0041] Furthermore, it should be noted that the present invention is not limited with respect to the raw sensor data to be evaluated. Rather, within the scope of the present invention, various raw sensor data generated during the operation of an application plant may be evaluated. For example, the following raw sensor data may be mentioned:

[0042] - Turbine speed of rotary sprayer,

[0043] - Air pressure of the driving air to drive the turbine of a rotary sprayer,

[0044] - Coating pressure of the paint pressure regulator,

[0045] - Charging current of the electrostatic coating agent charging system,

[0046] - Charging voltage of the electrostatic coating agent charging system,

[0047] - Humidity of the coating booth,

[0048] - Air pressure of molding air for spray formation of coating agent,

[0049] - Flow rate of molding air for forming a coating spray,

[0050] - Air temperature in the coating booth,

[0051] - Location of the paint impact point of the application device,

[0052] - Moving speed of the paint impact point of the application device,

[0053] - Valve, in particular, the valve position of a coating agent valve, flushing agent valve, pulse air valve, or lubricant valve,

[0054] - In particular, drive variables of the drive such as position, speed, acceleration, current, voltage, power, or temperature,

[0055] - Flow rate of the coating agent pump or dosing unit,

[0056] - Position of the linear conveyor conveying the component to be coated via the applicator,

[0057] - Material flow rate, temperature, and pressure,

[0058] - Speed ​​of the vortex applicator.

[0059] It has already been mentioned above that various raw sensor data can be evaluated as part of the monitoring process according to the present invention. In monitoring these different raw sensor data, any combination of the above examples of raw sensor data is possible. For example, the rotational speed of the turbine of a rotary sprayer can be evaluated together with the flow rate of the coating agent, which is just one example. Furthermore, any combination of the above raw sensor data is possible, and it is also possible within the scope of the invention to evaluate two or more different types of raw sensor data.

[0060] For example, raw sensor data can reflect the operating variable of one of the following components of an application plant:

[0061] - Motor,

[0062] - Robot joint of a coating robot,

[0063] - Drive controller of the robot drive of the coating robot,

[0064] - Turbine for driving a rotary sprayer,

[0065] - A molding air controller for controlling the molding air flow by molding the spray of a sprayer,

[0066] - Pumps, especially coating pumps,

[0067] - A metering device for measuring coating agents,

[0068] - Valves, especially proportional valves,

[0069] - Paint pressure regulator,,

[0070] - High-voltage generator for picking up electrostatic coating agents,

[0071] - Switch,

[0072] - Sensor,

[0073] - heating,

[0074] - Control system,

[0075] - Electric fuse,

[0076] - Electric battery,

[0077] - Uninterruptible Power Supply,

[0078] - Uninterrupted signal transmission,

[0079] - Transformer,

[0080] - Fluids, especially coating agents, viscous materials, adhesives, diluents, fluid lines for transporting air or water, especially hoses or pipes,

[0081] - Vortex applicator.

[0082] Regarding the provision of raw sensor data, various possibilities exist within the scope of the present invention. For example, raw sensor data may be measured continuously and then evaluated. Alternatively, if raw sensor data is measured earlier with a time delay, raw sensor data may be read from a database.

[0083] As mentioned above, features extracted from raw sensor data serve as a database for machine learning algorithms. These machine learning algorithms typically use so-called rules applied to the features to allow for inference regarding the operating status of the plant. To create these rules, previously measured and stored raw sensor data, along with so-called labels reflecting the operating status of the application plant (e.g., defect-free or defective) during past raw sensor data measurements, may be used. Then, features are extracted from the past raw sensor data in the manner described above. Subsequently, rules are determined through machine learning algorithms from features extracted from the past raw sensor data on one hand, and from known labels on the other. Rules determined in this manner can then be applied to features extracted from the currently measured raw sensor data.

[0084] The evaluation of features extracted from currently measured raw sensor data can be used, for example, to identify the operating status of an application plant. For example, one of the following conditions can be detected:

[0085] - Wear or defect in the paint pressure controller,

[0086] - Wear or defects in the mixer that mixes the various components of the coating agent,

[0087] - Wear or defects in the pump, especially the coating pump,

[0088] - Wear or defects in valves, especially those controlling coatings, viscous substances, adhesives, thinners, air or water,

[0089] - Wear or defect of the heater,

[0090] - Wear or defect in the drive motor,

[0091] - Electrical contact defect,

[0092] - Detection of air pockets in the paint or interruption of paint supply,

[0093] - In particular, the characteristics of the working medium of the application plant, such as air, water, varnish, adhesive, or viscous substances,

[0094] - Evaluation of applications and movement programs for stress on machine components or equipment,

[0095] - Detection of discharge of the bell cup of a rotary sprayer,

[0096] - Detection of contamination and / or moisture on the sprayer,

[0097] - Detection of abnormalities in the operating behavior of motors, pumps, pistons, molding air, fans, turbines, high voltage, and pressure flow regulators,

[0098] - Overall detection of abnormal signs in the sense of a significant deviation of the signal curve from the normal curve,

[0099] - Prediction of maintenance intervals.

[0100] Furthermore, the term "application plant" as used in the context of the present invention should be understood in a general sense and should be noted to include, among other things, a painting plant for painting automotive body components. However, the present invention may also be used in systems for, for example, adhesion, sealing, or insulation. Accordingly, the concept of a coating agent as used in the context of the present invention should be understood in a general sense and includes, among other things, paints and viscous materials such as adhesives or insulating materials.

[0101] Furthermore, it should be noted that the present invention does not claim protection only for the monitoring method according to the invention described above. Rather, the present invention also claims protection for the application plant, which comprises an applicator (e.g., a rotary sprayer), a manipulator (e.g., a multi-axis painting robot), a sensor for measuring raw sensor data, and a control device for controlling the application plant. The application plant according to the present invention is characterized in that the control device executes the monitoring method according to the invention described above.

[0102] In this sense, it should be noted here that control units can be distributed across multiple hardware components. It should also be noted that they do not necessarily have to be application plant control. Monitoring may be executed on completely different hardware entirely unrelated to application plant control.

[0103] Other advantageous additional embodiments of the present invention are set forth in the dependent claims or are described in more detail below with reference to the drawings, together with the description of preferred embodiments of the present invention. The drawings represent the following: Brief explanation of the drawing

[0104] FIG. 1 is a schematic diagram of a system structure according to the present invention for monitoring an application plant, Figure 2 is a schematic diagram of the structure of the analysis software, Figure 3 is a schematic diagram of the operating mode of the analysis software for feature extraction, FIG. 4 is a flowchart illustrating a monitoring method according to the present invention, Figure 5 is a timing diagram to explain feature extraction, FIG. 6 is a flowchart illustrating a monitoring method according to the present invention, and also Figure 7 is a time diagram showing feature extraction considering multiple control signals and multiple raw sensor data. Specific details for implementing the invention

[0105] FIG. 1 shows the system structure of a system according to the present invention, which includes a machine (1), a controller (2) with built-in analysis software (3), analysis hardware (4) with analysis software (5), visualization (6), and hardware (7) belonging to an external system.

[0106] In this embodiment, the machine (1) is a painting plant for painting a car body component having all the components.

[0107] Here, the controller (2) outputs a control signal to the machine (1), receives raw sensor data from the machine (1), and the raw sensor data can be analyzed by the analysis software (3).

[0108] Additionally, raw sensor data can also be evaluated by analysis software (5) of analysis hardware (4).

[0109] Figure 2 shows a schematic diagram of the modular structure of the analysis software (3 or 5).

[0110] Therefore, the analysis software (3) can access the data memory (8) and gateway (9) to communicate with other components such as the machine (1) or hardware (7).

[0111] Additionally, the analysis software (3) includes a feature extraction module (10), a rule generation module (11), and a rule application module (12).

[0112] FIG. 3 is a schematic diagram illustrating a monitoring method according to the present invention.

[0113] First, the feature extraction module (10) extracts features from past raw data (13) previously measured in a known operating state of the machine (1), where the known operating state of the machine (1) is indicated by a label (14). The feature extraction module (10) extracts features (15) from past raw sensor data (13), which will be described in detail. Since the feature extraction by the module (10) has the task of reducing the data volume of the past raw sensor data (13), a machine learning algorithm can subsequently be based on the features (15).

[0114] Module (11) uses extracted features (15) and known labels (14) to generate rules used to evaluate the currently measured raw data (16) later.

[0115] In the course of normal operation, the module (10) extracts features again from the new raw sensor data (16) currently measured, and the module (12) determines the state (17) of the machine (1) by applying a predetermined rule to the extracted features.

[0116] Next, the flowchart according to Fig. 4 is explained with reference to the time diagram according to Fig. 5.

[0117] In the first step (S1), the application plant is controlled, and raw signals are stored therein, namely, control signals for controlling the application plant on one hand and actual signals (raw sensor data) on the other, which are measured by the sensor and represent the operating variables of the application plant.

[0118] The raw signals (control signals and actual signals) are stored in step S2.

[0119] In step S3, previously defined parameters of each application are retrieved. Meanwhile, an observation time window [T, which could be, for example, a car body cycle. A , T EIt includes ]. On the other hand, the retrieved parameter also defines the selection of the desired signal set, namely the control signal and the actual signals to be evaluated. Finally, the retrieved parameter also includes the adjusting steps d DYN Number of n, controlling steps d DYN Duration of and controlled phase d STAT Includes parameters for dividing the comparison period [t1, t2], such as the duration of, into subsections. Adjusting steps d DYN Static controlled steps d that are still suitable for the comparison period after excluding STAT There is.

[0120] In step S4, the pre-stored data of the raw signals (control signals and actual signals) is the observation time window [T determined in step S3 A , T E ] and signal sets are specifically searched.

[0121] In step S5, the change in the considered control signal is determined, and by this, such a change occurs in the time diagram according to Fig. 5 at times t1 and t2.

[0122] Next, in step S6, at least one comparison time period [t1, t2] is determined, starting with a change in the control signal at time t1 and ending with the next change in the control signal at time t2. Generally, several comparison periods are considered.

[0123] In step S7, this comparison period [t1, t2] is divided into subsections according to the parameters retrieved in step S3. The duration of the comparison period [t1, t2], adjusting step d DYN Duration of and controlled phase d STAT Depending on the duration of, the comparison period [t1, t2] is 0 ... n adjusting step d DYN , 0 ... m adjustable step d STATand is divided into 0 ... 1 residual steps. In the time diagram according to Fig. 5, the comparison period (t1, t2) consists of three controlling steps d DYN and two regulated steps d STAT It consists of. However, this is just an example, and the comparison period [t1, t2] can also be composed of other types of subsections.

[0124] In step S8, the univariate statistical characteristics of the raw sensor data are determined for each individual section, that is, for the actual signals measured by the sensor and representing the operating variables of the application plant. The time diagram according to Fig. 5 specifically shows that the molding air flow of the rotary atomizer is measured as the actual signal. However, depending on the set of signals retrieved in step S3, statistical parameters may also be calculated using other actual signals. For example, this univariate statistical parameter may be an arithmetic mean, to give just one example.

[0125] Then, in step S9, features are formed using the univariate statistical features calculated in step S8, and said features are the components of these features. Other components of the features are the target values ​​and the changes in the target values.

[0126] In step S10, features are used to determine the state of the application plant, which can be performed by simple (passively parameterized) rules or machine learning algorithms.

[0127] Figure 6 shows a flowchart illustrating the measurement of raw sensor data in step S1, the extraction of features in step S2, and the application of rules to the extracted features in step S3.

[0128] Figure 7 shows a variation of the time diagram according to Figure 5, with two control signals and two actual signals displayed. The comparison period is limited by the jump of the two control signals.

[0129] The first comparison period is between times t1 and t2. The second comparison period is between t2 and t3.

[0130] The aforementioned comparison period can be further divided into sub-sections in the manner described above.

[0131] In addition, statistical parameters may be multivariate parameters calculated from, for example, two actual signals, but are recalculated for individual subsections. An example of such a multivariate statistical parameter is the Pearson correlation coefficient.

[0132] The present invention is not limited to the preferred embodiments described above. Rather, the concept of the present invention may be used, and thus many variations and modifications are possible within the scope of protection. In particular, the present invention also claims protection for the features and subject matter of the dependent claims, apart from the claims mentioned in each case, particularly without the features of the main claims. Accordingly, the present invention includes different aspects of the present invention that provide protection independently of one another. Explanation of the symbols

[0133] 1 machine 2 controllers 3 Analysis Software 4 Analysis Hardware 5 Analysis Software 6 Visualization 7 Hardware 8 Data Memory 9 Gateway 10 Feature Extraction Module 11 Rule Generation Module 12 Rule Application Modules 13 Past raw data 14 labels 15 Features 16 New raw data 17 Condition of the machine d DYN Adjustment stage dSTAT Controlled stage [T A , T E ] Observation Time Window [t1, t2] Comparison period

Claims

Claim 1 A monitoring method for an application plant (1) for applying a coating agent comprises the following steps: a) a step of determining first raw sensor data (16) representing operating variables of the application plant (1); and b) a step of acquiring a first control signal for controlling the application plant (1); the following steps: c) a step of extracting a feature (15) from the first raw sensor data (16), wherein the feature includes a reduced amount of data compared with the first raw sensor data (16) and serves as a data basis for a machine learning algorithm; and d) an observation time window ([T A , T E The method is characterized by the step of determining the timing of a change of a first control signal within ]) and the step of determining at least one comparison period ([t1, t2]) between two consecutive changes of the first control signal, and the following steps for extracting the features (15): c-1) an observation period window ([T) for evaluating the first or second raw sensor data (16). A , T E Step defining ]) c-2) Observation period window ([T A , T E Step c-3) defining at least one comparison period ([t1, t2]) within ]), dividing the comparison period ([t1, t2]) into several consecutive subsections (d DYN , d STAT , d REST ) a step of subdividing into, and c-4) the first or second raw sensor data (16) or individual subsections (d DYN , d STAT , d REST Characterized by the step of calculating at least one statistical parameter that is a component of the above feature for additional raw sensor data within ), wherein the comparison period is one or more initial adjustment steps (d) in which the first raw sensor data responds to a control signal step, or none at all. DYN ),e-2) regulated steps that follow in the order of occurrence in which the first raw sensor data has at least partially completed responding to the control signal step, is absent, or has one or more (d STAT ), and e-3) subsequent remaining steps in the order of occurrence until the end of the comparison period (d REST A monitoring method characterized by being divided into at least one of ) Claim 2 A monitoring method according to claim 1, characterized by the following steps: a) determining second raw sensor data (16) that reflects operating variables of an application plant (1) different from first raw sensor data (16); b) extracting from the second raw sensor data (16) a feature (15) that includes a reduced amount of data compared to the second raw sensor data (16) and functions as a data basis for a machine learning algorithm. Claim 3 delete Claim 4 delete Claim 5 A monitoring method according to claim 1, characterized by the following steps: a) detecting a second control signal for controlling an application plant (1); b) determining the change time of the first control signal within an observation time window; c) determining the change time of the second control signal within an observation time window; and d) determining at least one comparison period between two consecutive changes of the two control signals. Claim 6 A monitoring method according to claim 1, characterized in that the following steps a) determining the amount of change of a first control signal or a second control signal, b) determining the time period after the last change of the first control signal or the second control signal, and c) the amount of change and the time period from the last change are part of the feature (15). Claim 7 delete Claim 8 A monitoring method according to claim 1, characterized by the following steps: a) determining a time constant of first raw sensor data (16); and b) determining at least one of the following quantities as a function of the time constant of the first raw sensor data (16): b1) adjusting a comparison period (d DYN Number of ), b2) Steps for individual adjustment (d DYN ) duration of, b3) individually controlled step (d STAT Duration of ) Claim 9 In claim 8, the individual adjustment steps (d DYN ) or individually controlled steps (d STAT A monitoring method characterized by the duration of ) being determined randomly. Claim 10 In claim 1, the observation time window ([T A , T E A monitoring method characterized by the following periods: a) a component-related time period, a1) a preparation period of the application plant for coating the component for subsequent coating, a2) a coating period of the component, b) a time period related to the application plant, b1) an installation and operation period of the application plant (1), b2) a test operation period of the application plant (1), b3) a manual operation period of the application plant (1), b4) a maintenance operation period of the application plant (1), b5) a production operation period of the application plant (1), b6) a time period for automated secondary processes related to the plant components (e.g., rinsing over time, automatic testing, or diagnostic measures in which the system component is specifically controlled), c) a time-related period, c1) one hour, one day, one week, one month, one quarter, one year, or c2) one of the shift lengths of the operation shift. Claim 11 A monitoring method according to claim 1, wherein the statistical parameter is one of the following parameters: a) a univariate statistical parameter considering only the first raw sensor data (16) below, a1) arithmetic mean value, a2) geometric mean value, a3) median value, a4) variance, a5) maximum value, a6) minimum value; b) a multivariate statistical parameter considering, in addition to the first raw sensor data (16), the second raw sensor data (16) and optionally the additional raw sensor data (16) below, b1) Pearson correlation coefficient, b2) rank correlation coefficient Claim 12 A monitoring method according to claim 1, characterized in that the first raw sensor data (16) or the second raw sensor data (16) or additional raw sensor data (16) reproduces one of the following operating variables of the application plant (1), or the first control signals or the second control signals control one of the following operating variables of the application plant (1): a) turbine speed of a rotary atomizer, b) air pressure of driving air for driving the turbine of the rotary atomizer, c) coating agent pressure of a paint pressure regulator, d) charging current of an electrostatic coating agent charging system, e) charging voltage of an electrostatic coating agent charging system, f) humidity of a coating booth, g) air pressure of molding air for forming a spray of coating agent, h) mass flow rate of molding air for forming a spray of coating agent, i) air temperature of a coating booth, j) position of a paint impact point of a coating device, k) moving speed of a paint impact point of a coating device, l) valve position of a valve, m) drive variable of a drive, n) coating agent pump or o) flow rate of the dosing unit, o) position of the linear conveyor conveying the component to be coated through the application plant (1), p) flow rate, temperature and pressure of the material, q) rotational speed of the vortex applicator Claim 13 A monitoring method according to claim 1, wherein the first raw sensor data (16) or the second raw sensor data (16) reflects an operating variable of one of the following components of an application plant (1): a) a motor, b) a robot joint of a coating robot, c) a drive controller of a robot drive of a coating robot, d) a turbine for driving a rotary sprayer, e) a molding air controller for controlling the flow of molding air to form a spray of a sprayer, f) a pump, g) a metering device for metering a coating agent, h) a valve, i) a paint pressure regulator, j) a high voltage generator for electrostatic coating agent charging, k) a switch, l) a sensor, m) a heater, n) a control system, o) an electric fuse, p) an electric battery, Q) an uninterruptible power supply, r) an uninterruptible signal transmission, s) a transformer, t) a fluid line for transporting fluid, u) a vortex applicator Claim 14 A monitoring method according to claim 1, characterized by the following steps for determining the first raw sensor data: a) measuring the first raw sensor data (16) with a first sensor, or b) reading the first raw sensor data (16) previously measured and stored from a database. Claim 15 A monitoring method according to claim 1, characterized by the following steps: acquiring rules through the following steps to establish rules to be used for subsequent evaluation of first raw sensor data: a) a step of retrieving stored past raw sensor data (13); b) a step of determining or retrieving a stored label (14) that is temporally associated with the past raw sensor data (13), wherein the label reflects the state of the application plant (1) when the past raw sensor data (13) is measured; c) a step of extracting features (15) from the past raw sensor data (13); and d) a step of determining rules through a machine learning algorithm by evaluating the features (15) determined from the past raw sensor data (13) and the associated label (14). Claim 16 A monitoring method characterized by the following steps in any one of claims 1 to 2, 5 to 6, and 8 to 14: a step of establishing rules for subsequent evaluation of initial raw sensor data by determining rules by the following steps: a) a step of searching for stored past raw sensor data; b) a step of extracting features from past raw sensor data; c) a step of determining rules through an unsupervised machine learning algorithm by evaluating features determined from past raw sensor data without associated labels. Claim 17 A monitoring method characterized by the following steps in any one of claims 1 to 2, 5 to 6, and 8 to 14: a step of determining rules by the following steps to generate rules for subsequent evaluation of initial raw sensor data; a) a step of searching for stored past raw sensor data; b) a step of extracting features from past raw sensor data; c) a step of determining rules through a human expert by evaluating features determined from past raw sensor data without associated labels. Claim 18 A monitoring method according to claim 1, characterized by the following steps: evaluating the characteristics (15) of the currently determined first raw sensor data (16) or second raw sensor data (16) using a machine learning algorithm for determining the operating state of the application plant (1) to detect one of the following operating states of the application plant (1): a) wear or defect of a paint pressure regulator, b) wear or defect of a mixer that mixes various components of a coating agent together, c) wear or defect of a pump, d) wear or defect of a valve, e) wear or defect of a heater, f) wear or defect of a drive motor, g) electrical contact defect, h) detection of air entrapment in the paint or interruption of paint supply, i) characteristics of the operating medium of the application plant (1), j) evaluation of the application and movement program for stress on the machine components or equipment, k) detection of bell plate discharge, l) detection of contamination or moisture on the sprayer, m) detection of anomalies in the operating behavior of the motor, pump, piston, steering gap, turbine, high voltage, and flow regulator, n) normal Overall detection of anomalies signifying a significant deviation of the signal curve from the curve, o) Prediction of maintenance intervals Claim 19 An application plant (1) for coating a part, comprising: a) at least one applicator for applying a coating agent; b) at least one manipulator for moving the applicator; c) at least one sensor for measuring first raw sensor data representing an operating variable of the application plant (1); and d) a control device (2-5) for controlling the application plant (1), wherein the control device (2-5) is arranged as a single component or distributed across a plurality of components; and e) the control device (2-5) or a separate monitoring device performs a monitoring method according to claim 1.

Images