Data Generator
Overview
The Data Generator in my8data enables the creation of synthetic test data for training, demonstration, and testing purposes. Instead of having to use real measurement data, you can use the data generator to systematically generate datasets with specific statistical properties.
Applications
| Application | Description |
|---|---|
| Training and Education | Create exercise data with known properties to practice interpreting MFU, SPC, and process capability |
| Demonstration | Show customers or colleagues the capabilities of my8data with realistic but anonymized data |
| Module Testing | Verify the correct function of analysis modules with data whose expected results are known |
| Method Comparison | Compare different statistical methods using identical datasets |
| Sensitivity Analysis | Investigate how changes to parameters (e.g., variation, mean) affect process capability indices |
Info: The generated data is created using a random generator and is suitable for statistical analysis. It does not represent real measurement values and should not be used for productive quality decisions.
Quick Start
Generate a test dataset in just a few steps:
- Open the Data Generator from the main menu
- Select the desired distribution type (e.g., normal distribution)
- Set the parameters (mean, standard deviation, sample size)
- Click Generate
- Import the generated data directly into an analysis module or export it
Tip: To test the impact of various scenarios, generate multiple datasets with different parameters. For example, you can simulate a capable process (Cpk >= 1.67) and an incapable process (Cpk < 1.00) and compare the results.
Parameters and Configuration
Distribution Types
The data generator supports various distribution types to simulate different process situations:
| Distribution Type | Description | Typical Application |
|---|---|---|
| Normal Distribution | Symmetric bell curve; most common in practice | Standard case for most manufacturing processes |
| Uniform Distribution | All values in range equally probable | Simulation of a process without clear central tendency |
| Log-Normal Distribution | Right-skewed; only positive values | Roughness, particle sizes, failure times |
| Weibull Distribution | Flexible shape; can be skewed or symmetric | Lifetime analysis, reliability |
Configurable Parameters
Normal Distribution
| Parameter | Description | Example Value | Effect |
|---|---|---|---|
| Mean (μ) | Center of the distribution | 10.00 | Shifts the entire distribution left or right |
| Standard Deviation (σ) | Width of the distribution | 0.02 | Larger values create wider spread |
| Sample Size (n) | Number of values to be generated | 100 | More values increase statistical significance |
Uniform Distribution
| Parameter | Description | Example Value |
|---|---|---|
| Minimum | Lower limit of the value range | 9.90 |
| Maximum | Upper limit of the value range | 10.10 |
| Sample Size (n) | Number of values to be generated | 100 |
Log-Normal Distribution
| Parameter | Description | Example Value |
|---|---|---|
| μ (log) | Mean of the logarithmized characteristic | 2.30 |
| σ (log) | Standard deviation of the logarithmized characteristic | 0.10 |
| Sample Size (n) | Number of values to be generated | 100 |
Weibull Distribution
| Parameter | Description | Example Value |
|---|---|---|
| Shape Parameter (k) | Determines the shape of the distribution | 3.5 |
| Scale Parameter (λ) | Characteristic lifetime / scaling | 10.00 |
| Sample Size (n) | Number of values to be generated | 100 |
Info: For the Weibull distribution, a shape parameter k < 1 produces a decreasing distribution (early failures), k = 1 corresponds to an exponential distribution (random failures), and k > 3 yields an approximately bell-shaped distribution (wear-out failures).
Advanced Options
In addition to the basic parameters, you can make the following advanced settings:
| Option | Description | Default Value |
|---|---|---|
| Seed | Starting value for the random generator; enables reproducible results | Random |
| Decimal Places | Number of decimal places | 3 |
| Add Outliers | Deliberately add outliers to the dataset | Disabled |
| Number of Outliers | How many outliers to insert | 0 |
| Outlier Range | Range in which the outliers should lie | ±4σ to ±6σ |
Tip: Use the Seed function when you need reproducible results. With the same seed and identical parameters, you will always get the same dataset. This is particularly useful for training where all participants should work with the same data.
Practical Examples
Example 1: Simulate a Capable Process
Goal: Generate a dataset with Cpk >= 1.67
| Parameter | Value | Rationale |
|---|---|---|
| Distribution | Normal Distribution | Standard case |
| Mean | 10.000 | Centered on target value |
| Standard Deviation | 0.010 | Low variation |
| Sample Size | 100 | Sufficient for Ppk calculation |
| USL | 10.050 | Tolerance width = 0.100 mm |
| LSL | 9.950 |
Expected Result: Cp ≈ Cpk ≈ 1.67 (Tolerance width 0.100 / 6 * 0.010 = 1.67)
Example 2: Simulate a Decentered Process
Goal: Generate a dataset with good Cp but poor Cpk
| Parameter | Value | Rationale |
|---|---|---|
| Distribution | Normal Distribution | Standard case |
| Mean | 10.025 | Deliberately off target |
| Standard Deviation | 0.010 | Same variation as Example 1 |
| Sample Size | 100 | |
| USL | 10.050 | |
| LSL | 9.950 |
Expected Result: Cp ≈ 1.67, but Cpk ≈ 0.83 (Process variation is narrow but decentered)
Warning: Synthetically generated data follows the selected distribution exactly. Real process data in practice often deviates from ideal distributions. Results achieved with generated data are therefore often "cleaner" than real analyses.