var_quantity

This crate is an extension of dyn_quantity and provides an interface for defining variable quantities whose value is a (pure) function of other quantities.

As an example, let's consider the eddy current losses in a conductive material which are caused by sinusoidally changing magnetic fields. A simple model could only take the magnetic flux density amplitude into account and a more sophisticated model would also consider the field frequency. Using the VarQuantity wrapper, both models can be used with the same interface:

use dyn_quantity::{DynQuantity, PredefUnit, Unit};
use var_quantity::{QuantityFunction, VarQuantity, FunctionWrapper};
use uom::si::{f64::{Power, MagneticFluxDensity, Frequency}, 
    power::watt, magnetic_flux_density::tesla, frequency::hertz};

// The serde annotations are just here because the doctests of this crate use
// the serde feature - they are not needed if the serde feature is disabled.

// Model 1: p = k * B^2
#[derive(Clone, serde::Deserialize, serde::Serialize)]
struct Model1(DynQuantity<f64>);

#[typetag::serde]
impl QuantityFunction for Model1 {
    fn call(&self, influencing_factors: &[DynQuantity<f64>]) -> DynQuantity<f64> {
        let mut b = DynQuantity::new(0.0, PredefUnit::MagneticFluxDensity);
        for factor in influencing_factors.iter() {
            if b.unit == factor.unit {
                b = factor.clone();
            }
        }
        return self.0 * b.powi(2);
    }
}

// Model 2: p = k * f^2 * B^2
#[derive(Clone, serde::Deserialize, serde::Serialize)]
struct Model2(DynQuantity<f64>);

#[typetag::serde]
impl QuantityFunction for Model2 {
    fn call(&self, influencing_factors: &[DynQuantity<f64>]) -> DynQuantity<f64> {
        let mut b = DynQuantity::new(0.0, PredefUnit::MagneticFluxDensity);
        let mut f = DynQuantity::new(0.0, PredefUnit::Frequency);
        for factor in influencing_factors.iter() {
            if b.unit == factor.unit {
                b = factor.clone();
            }
            if f.unit == factor.unit {
                f = factor.clone();
            }
        }
        return self.0 * f.powi(2) * b.powi(2);
    }
}

let k = DynQuantity::new(
    1000.0,
    Unit::from(PredefUnit::Power) / Unit::from(PredefUnit::MagneticFluxDensity).powi(2),
);
let model1: VarQuantity<Power> = VarQuantity::Function(
    FunctionWrapper::new(Box::new(Model1(k))).expect("output unit is watt"),
);

let k = DynQuantity::new(
    2.0,
    Unit::from(PredefUnit::Power)
        / Unit::from(PredefUnit::MagneticFluxDensity).powi(2)
        / Unit::from(PredefUnit::Frequency).powi(2),
);
let model2: VarQuantity<Power> = VarQuantity::Function(
    FunctionWrapper::new(Box::new(Model2(k))).expect("output unit is watt"),
);

// This function takes a variable quantity, the magnetic flux density and
// the frequency and calculates the losses
fn losses(model: &VarQuantity<Power>, b: MagneticFluxDensity, f: Frequency) -> Power {
    return model.get(&[b.into(), f.into()]);
}

let b = MagneticFluxDensity::new::<tesla>(1.2);
let f = Frequency::new::<hertz>(20.0);

assert_eq!(losses(&model1, b, f).get::<watt>(), 1440.0);
assert_eq!(losses(&model2, b, f).get::<watt>(), 1152.0);

The workflow to use the interface of this crate is as follows:

• Define the relation between input and output by implementing QuantityFunction for the type representing a variable quantity (Model1 and Model2 in the previous example). The implementor is responsible for selecting the right quantities for his model from the give influencing_factors (for unary functions, the crate provides filter_unary_function to simplify this) and also for defining sensible defaults if the needed quantity is not given (in the example above, the default flux density and frequency was defined to zero). As explained in the serialization / deserialization section, the types must not be generic.
• Create an type instance and box it as a trait object. The trait object approach is necessary for two reasons:
  1. Reduce generic bloat (for example, when a material type is defined using multiple VarQuantity for different properties, this could lead to dozens of generic parameters).
  2. To allow for serialization and deserialization using the typetag crate.
• Wrap the trait object in a FunctionWrapper. Since QuantityFunction works with dynamic quantities, it needs to be tested whether the output from QuantityFunction::call can be converted to the statically typed quantity T using TryFrom<DynQuantity<f64>> (in the example, the quantity types provided by the uom crate were used). This check is done in the constructor FunctionWrapper::new and again in FunctionWrapper::call, see the docstring of FunctionWrapper.
• Wrap the FunctionWrapper in VarQuantity::Function. The purpose of this enum is to offer an optimization for the important case of a constant quantity via its second variant VarQuantity::Constant. Its VarQuantity::get method either returns the constant quantity directly or forwards to FunctionWrapper::call.

Predefined variable quantity models

Some variable quantity models are very common and therefore provided with this crate. For example, model 1 from the introduction could also be realized using the Polynomial struct from the unary module:

use dyn_quantity::{DynQuantity, PredefUnit, Unit};
use var_quantity::{unary::Polynomial, VarQuantity, FunctionWrapper};
use uom::si::{f64::{Power, MagneticFluxDensity, Frequency}, 
    power::watt, magnetic_flux_density::tesla, frequency::hertz};

// The input vector [a, b, c] is evaluated as ax² + bx + c. Here, b and c are
// zero, but still need to match unit-wise:
// [a] = W/T², [b] = W/T, [c] = W
// The output unit is [c] and the input unit is calculated as [c/b].
// [a] (and additional terms) can then be checked.
let a = DynQuantity::new(1000.0, Unit::from(PredefUnit::Power) / Unit::from(PredefUnit::MagneticFluxDensity).powi(2));
let b = DynQuantity::new(0.0, Unit::from(PredefUnit::Power) / Unit::from(PredefUnit::MagneticFluxDensity));
let c = DynQuantity::new(0.0, PredefUnit::Power);
let polynomial = Polynomial::new(vec![a, b, c]).expect("terms are checked during construction");

let model1: VarQuantity<Power> = VarQuantity::Function(
    FunctionWrapper::new(Box::new(polynomial)).expect("output unit is watt"),
);

// This function takes a variable quantity, the magnetic flux density and
// the frequency and calculates the losses
fn losses(model: &VarQuantity<Power>, b: MagneticFluxDensity, f: Frequency) -> Power {
    return model.get(&[b.into(), f.into()]);
}

let b = MagneticFluxDensity::new::<tesla>(1.2);
let f = Frequency::new::<hertz>(20.0);

assert_eq!(losses(&model1, b, f).get::<watt>(), 1440.0);

For a full list of available models, see the following modules:

• unary : Models representing unary functions (single input).

Serialization and deserialization

The serde integration is gated behind the serde feature flag.

All structs / enums in this crate implement serialization and deserialization. See the docstrings of the individual types for details. The trait objects stored within FunctionWrapper are handled via typetag, which is why the the implementors of QuantityFunction cannot be generic.

Documentation

The full API documentation is available at https://docs.rs/var_quantity/0.1.0/var_quantity/.